Datasets:

babytreecc

/

ICLR2025review

like 7

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "-\tThe paper represents subtrees in question code using simple one-hot encoding without additional processing, which might ignore the frequency of subtrees and their structural relationships. Additionally, this representation can be very sparse. How does this affect CodePlexity's performance?\n-\tTo what extent does the choice of code generation model affect the method's results? Would using different code generation models lead to identifying different patterns of complexity, and how might this impact the method's consistency?\n-\tSince the method evaluates difficulty without considering visual information, there might be cases where code generation is challenging due to question ambiguity, but the task would be straightforward with visual context. Is CodePlexity truly measuring question difficulty, or is it primarily measuring code generation complexity?\n-\tMost models evaluated in the paper are large video-language models built upon language model architectures. Since these models share similar language model foundations with the code generation approach, they might inherently struggle with the same types of questions that are difficult for code generation. Does this architectural similarity explain their correlated performance? How well would the method generalize to fundamentally different architectures that might employ different reasoning mechanisms?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "-\tThe paper introduces a novel and interesting approach using code complexity to evaluate question complexity.\n-\tThe paper offers interesting insights into the differences in human evaluation, text-based and code-based evaluation.\n-\tThe experiment results show clear empirical evidence to support the claims." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces an approach to analyzing and generating complex questions for Video QA. The authors propose the use of generated code complexity as a metric to evaluate question difficulty. They introduce \"CodePlexity,\" a novel algorithm that analyzes generated code's structure and content to identify patterns that make questions challenging for ML models. This allows the creation of a new VideoQA dataset named \"CodePlex-QA\" which features more complex questions than existing datasets without relying on human expertise." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Please see the question section below" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Why the authors do not validate the proposed approach on other VideoQA datasets besides NExT-QA?\n- How do errors in code generation impact the complexity metrics?\n- Are there any important patterns that might be missed by your current subtree merging approach?\n- What specific criteria were used to manually filter out the 12% of questions?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- The idea to measuring question complexity using code generation is well-motivated\n- The ablation studies and analysis are well-designed" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a novel approach to analyzing and measuring question complexity in VideoQA by leveraging generated code complexity. The authors demonstrate that code-based metrics correlate better with model performance compared to human judgment and introduce CodePlexity, an algorithm for estimating question complexity. They use this to identify challenging patterns in VideoQA and automatically generate a new benchmark dataset, CodePlex-QA, which proves to be significantly more challenging than existing datasets." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- All experiments are conducted on a single dataset (NExT-QA) for analysis. Therefore, it leads to the limited evaluation across different types of VideoQA tasks.\n- The authors should discuss more how errors of code generation might affect the complexity metrics\n- The human evaluation study uses only 150 questions, which is a relatively small sample\n- We concern that the merging of subtrees that \"always co-occur\" could potentially miss important patterns in edge cases\n- The manual filtering process for the generated dataset (removing 12% of questions) could introduce selection bias" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. What are the results of the recent visual programming methods on this task?\n2. How complex is CodePlex-QA compared with Next-QA ATP-T/ATP-C split?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper proposes a novel metric to evaluate the complexity of the VideoQA problem and also proposes a CodePlex-QA that is more difficult for current VideoQA models. The idea of leveraging code to develop a new metric for question difficulty analysis is interesting.\n2. The paper conducts thorough experiments on testing the correlation of the metric." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the author claims that the questions considered difficult are different from the actual difficult questions to current VideoQA models. Therefore, the authors propose a new metric: CodePlexity, to evaluate the difficulty of a VideoQA question. This metric is based on the recent visual programming methods, where a visual program of a question is generated and which complexity is evaluated. Based on this metric, the authors found that most models struggle with questions where multiple frames are included and frame order need to be take into consideration. Then based on the metric, the author proposes CodePlex-QA and claims that it is 1.9 times harder than existing benchmarks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The criteria for what makes a question complex or challenging for models—especially compared to human intuition—seem speculative. Without more rigorous validation, it’s unclear whether the proposed complexity measures truly capture inherent question difficulty or just reflect the limitations of current VideoQA models. Also, the idea of identifying questions that are “easy for humans but hard for machines” is ambiguous. It seems plausible that any difference in difficulty may be more a result of model architecture and training rather than the intrinsic complexity of the question itself.\n2. Visual programming is a natural way (and probably the best way) to address the CodePlex-QA task. The author didn't report how well the recent visual programming methods (Visprog, ViperGPT, CodeVQA), especially those on addressing complex logical questions (eg. RVP[1]) addresses the task.\n3. The comparison between Next-QA and CodePlex-QA (Table2) is not convincing enough as previous works have shown that Next-QA have easy questions[2]. How is CodePlex-QA compared with Next-QA ATP-Hard split?\n\n[1] Recursive Visual Programming\n[2]Revisiting the \"Video\" in Video-Language Understanding" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- SeViLA \"leverages a single image-language model (BLIP2) to tackle both temporal keyframe localization and question answering on videos\" How the findings in this paper generalize to models that take multiple video frames for VQA? (e.g., VideoChat2 [5], Video-LLaVA [6])?\n\n- NExT-QA uses videos from VidOR [7] -- CodePlex-QA use videos from MOMA, ActivityNet, ActivityNet-Entities, ActivityNet-Captions), Action Genome, and Charades. Why not using the same videos, or at least the same source video dataset (VidOR)?\n\n[5] Li, Kunchang, et al. \"Mvbench: A comprehensive multi-modal video understanding benchmark.\" Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.\n\n[6] Lin, Bin, et al. \"Video-llava: Learning united visual representation by alignment before projection.\" arXiv preprint arXiv:2311.10122 (2023).\n\n[7] Xindi Shang, Donglin Di, Junbin Xiao, Yu Cao, Xun Yang, and Tat-Seng Chua. Annotating objects and relations in user generated videos. In ICMR, pages 279–287, 2019" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "+ This paper is easy to read, the method is easy to follow and the qualitative examples help to illustrate the motivation of their work.\n\n+ Leveraging Visual Programming outputs as a proxy for assessing the difficulty of a given task is an underexplored domain -- in software engineering, an active area of study is the correlation between the code complexity and difficulty of the task. This is an open problem, and this paper proposes an interesting generalization of this task.\n\n+ Decomposing the output code from the VP module via ASTs, and identifying subtrees that decrease performance via scoring, may give an interesting angle for generating an interpretable pipeline to analyze subprocesses that might be hurting the models' performance. In principle, this looks like an interesting approach and a viable metric for VP-based methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper explores a data-driven method for evaluating question complexity for Video Question Answering (VQA) by collecting the visual programs generated by ViperGPT. Then, they analyze the code outputs and leverage them to assess the difficulty of the questions in a VQA triplet.\nGiven the output code from the Visual Programming module (ViperGPT), they propose CodePlexity, an algorithm that parses the generated code via Abstract Syntax Trees (AST), and later identifies valid subtrees. These subtrees are then scored by correlating subtrees with difficult questions -- subroutines that hurt the VQA model performance. The authors use NExT-QA as the benchmark to analyze their proposed metric. \nGiven this metric, they also propose a pipeline to generate difficult question-answer pairs for a set of videos from MOMA, ActivityNet (+Entities/Captions), Action Genome, and Charades. \nFinally, they compare the results of models SeViLA-ZS, InternVideo and VIOLET on their proposed new dataset (CodePlex-QA) against NExT-QA." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Visual Programming (VP) is an interpretable framework for several visio-linguistic tasks, however, VP may output very simple code for a quite complex task, yielding false positive or false negative hard cases, and VP might not be reliable -- for a very complex task, the output code could be simple; the LLM might regard it as a simple task, but it's actually the opposite\nIn addition, VP falls short against end-to-end models (e.g., ViperGPT acc. is 60% vs SeViLa (finetuned) acc. is 73.8%) -- given its underperformance, it is hard to justify the use of its outputs as a proxy for evaluation of complexity. It is also hard to justify using VP for measuring task complexity and then evaluating on end-to-end models.\n\n- Disregarding visual inputs: \"Text-based metrics (above) perform worse than the code-based ones (below), and our approach demonstrates the highest correlation with the models’ performance.\" -- this completely ignores the visual modality, which seems problematic.\n\n- The authors claim: \"In summary, we discovered that VideoQA methods struggle with fine-grained temporal reasoning and lack spatio-temporal, object-centric representations. This is in accord with prior studies\" [3] -- however, those studies regard both visual and language modalities for assessing temporal reasoning in video QA, giving high importance to the video part.\n\n- Human evaluations: Figure 6: We ask human annotators to provide the relative ordering of three provided questions\naccording to the estimated complexity of answering the question about an unseen video -- how to correctly assess the complexity of the task without looking at the video?\n\n- Experimental setup:\nBaselines [1] focus on grammar and text only. BERT and GPT-4 also focus on text only.\n\n- Generalizability: ViperGPT is the VP module used in this paper, however, VP has significantly progressed. Different methods leverage multiple LLMs, there has been extensive problem decomposition and iteration that impact the code outputs. Further experiments with other VP approaches might be required to ensure generalization. Similarly, the proposed metric and dataset is only compared against NExT-QA. Other benchmarks might be necessary to validate the proposed metric and dataset (e.g., MVBench [2] compiles a collection of datasets (with subsets of samples from a diverse range of sources), which includes spatio-temporal analysis, spatial action, object, position, scene, count, attribute and cognition).\n\n- For dataset generation, the authors compare their pipeline with [4] -- however, an important step for EgoSchema is the manual curation of the whole dataset in the final step -- details for this step is not further detailed in this paper.\nFurthermore, as a comparison, EgoSchema consists of over 5000 human curated multiple choice question answer pairs, spanning over 250 hours of real video data -- significantly larger than the 1981 samples -- what is the length of each video sample?\n\n\n[1] Yulia Tsvetkov, Manaal Faruqui, Wang Ling, Brian MacWhinney, and Chris Dyer. Learning the curriculum with bayesian optimization for task-specific word representation learning. In ACL, 2016. \n\n[2] Li, Kunchang, et al. \"Mvbench: A comprehensive multi-modal video understanding benchmark.\" Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.\n\n[3] Shyamal Buch, Crist´obal Eyzaguirre, Adrien Gaidon, Jiajun Wu, Li Fei-Fei, and Juan Carlos Niebles. Revisiting the” video” in video-language understanding. In CVPR, 2022.\n\n[4] Karttikeya Mangalam, Raiymbek Akshulakov, and Jitendra Malik. EgoSchema: A diagnostic benchmark for very long-form video language understanding. In NeurIPS, 2023." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We propose a data-driven method to assess question complexity in Video Question Answering (VideoQA) based on code primitives, and use it to create a new benchmark, which is nearly twice as difficult for models compared to existing datasets." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024understanding,\ntitle={Understanding Complexity in Video{QA} via Visual Program Generation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0yXqV8VJKi},\nnote={under review}\n}" }, "abstract": { "value": "We propose a data-driven approach to analyzing query complexity in Video Question Answering (VideoQA). Previous efforts in benchmark design have largely relied on human expertise to construct challenging samples. In this work, we experimentally demonstrate that humans struggle to accurately estimate which questions are hard to answer for machine learning models. \n Our alternative, automated approach takes advantage of recent advances in code generation for visual question answering. In particular, we use generated code complexity as a proxy for the question complexity and demonstrate that it indeed shows a much stronger correlation with the models' performance, compared to human estimates. We then present a novel algorithm for estimating question complexity from code. It identifies fine-grained primitives which correlate with the hardest questions. These human-interpretable results lead to a number of discoveries about the key sources of complexity for VideoQA models. Finally, we extend our approach to generate complex questions for a given set of videos. This allows us to automatically construct a new benchmark, which is 1.9 times harder for VideoQA methods than existing manually designed datasets." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "video understanding", "codegen" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/5dca235cf7060cf72798cff5a5c86564629f69c8.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/0b2646320259785d00546826ecb9443f8ba683c1.zip" }, "title": { "value": "Understanding Complexity in VideoQA via Visual Program Generation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper is well-written and well-organized. The ideas are clear and could be easily understood\n2. The experiments are comprehensive and the results are strong. As shown in Section 4, the proposed BTR algorithm could greatly outperform state-of-the-art baselines in two classic benchmarks and handle three hard and complex modern games with a desktop PC.\n3. The paper includes extensive ablation studies and experimental data. Section 5 presents a detailed analysis of the performance and impact of each component of the BTR algorithm, providing readers with insights into the sources of the algorithm's performance gains. Additionally, the authors include complete experimental results and settings in the appendix, helping to clarify any potential confusion or misunderstanding for readers." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces Beyond The Rainbow (BTR), a novel reinforcement learning (RL) algorithm that enhances Rainbow DQN by integrating six key improvements. The BTR algorithm is computationally efficient, capable of training powerful agents on a standard desktop computer within a short time. Experimental results show that BTR outperforms state-of-the-art RL algorithms on both the Atari-60 and Procgen benchmarks. Additionally, BTR can handle training agents for challenging levels in complex, modern games. Finally, this paper includes a comprehensive ablation study to analyze the performance and impact of each component within the BTR algorithm." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The BTR integrates six improvements from existing RL literature to Rainbow DQN. While the algorithm demonstrates strong performance, its novelty might appear limited. Could you further clarify the novelty of this work? Or specifically, could you briefly discuss if there is any challenges in integrating these existing improvements into the BTR algorithm?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "I am just slightly worried about the use of somewhat modern Nintendo games as RL environments through the use of emulators, is the use of emulators for research legal?" }, "flag_for_ethics_review": { "value": [ "Yes, Legal compliance (e.g., GDPR, copyright, terms of use)" ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. What exactly is adaptive maxpooling? Would it be possible to add a description of it with either an equation, pseudo-code, or diagram?\n2. Where did the formula 0.05/batch_size for Adam's epsilon come from?\n3. The final algorithm has a considerable number of hyperparameters, would it be possible to discuss a bit which ones are the most important to tune should someone try to apply this algorithm to a new domain?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The presentation is overall clear, the methodology is sound, and the results are compelling. Both extensive use of ablations, and the connection to other important metrics related to pathologies in Deep RL algorithms are an example that more papers should follow.\nThe appendices are also data rich, showing ablations' performances on each of ALE's 60 games, and even having one appendix about things that were tried but did not lead to improvements in performance, which may help others not repeat the experiments." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents a variant of Rainbow that adds further architectural and algorithmic improvements to improve not only the agent's score but also to increase its training speed to around 3x what has been previously reported, while running it on top-notch consumer hardware. Finally the authors also show that their improved version of rainbow can deal with modern games with complex graphics and physics." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Adaptive Maxpooling is never defined. It's not a common layer in reinforcement learning and it's never defined in the paper, in fact skimming (Schmidt and Schmied, 2021) that layer is also not defined, I believe this is the only seious weakness in the paper's presentation, but still I believe it is a serious weakness (though hopefully the authors can fix it and so I can increase their grade).\n2. There are at least 2 relevant citations missing, \"Spectral Normalisation for Deep Reinforcement Learning: An Optimisation Perspective\" when talking about Spectral Normalisation, and \"On the consistency of hyper-parameter selection in value-based deep reinforcement learning\" when talking about the need for tuning Deep RL hyperparameters and the benefits of using layer norm between dense layers.\n3. I believe it's slightly misleading to not specify \"a high-end PC\" when talking about the kind of machine that can run the algorithm in 12 hours (4090 RTXs are quite expensive, and i9s are Intel's high-end consumer line)\n4. I believe a more direct comparison with Schmidt and Schmied, 2021 is warranted, given its foundational importance to the paper.\n5. Using only 3 seeds while having a large increase in the number of tuned hyperparameters weakens the validity of the results as explained in \"Empirical Design in Reinforcement Learning\", though at the same time the analysis of metrics beyond simply the score and the extensive use of ablations help." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- The 12 hour number/other timings, is that the total time it takes to train BTR on a single game or on all 57 games?\n- It seems like you made quite a few hyperparameter choices (e.g. how often to update, etc.) Do you use the same values for each domain?\n- What is the shaded area in the plots? If it is standard deviation it seems like the proposed BTR algorithm is very inconsistent across seeds. Could you elaborate please/maybe provide results for individual seeds?\n- Figure 3 does show that you can apply your approach to other games, which is great. I would really like to see some point of comparison, however, to act as a reference point. For instance, run vanilla PPO or DQN or Rainbow as a baseline.\n- Why is fig 4 using raw score as the y-axis, as opposed to e.g. normalised?\n- Figure 4 is somewhat hard to follow as there are so many lines and it seems like most of them overlap quite a lot.\n- Is it feasible to run rainbow with vectorisation? This is not that crucial, it just seems like something obvious to run given figure 5, where vectorisation is the main speedup factor.\n- Table 2: Would be nice to have another method, e.g. rainbow or DQN to act as a reference point.\n- One recent work that seems to have a similar purpose is \"Simplifying Deep Temporal Difference Learning\" (https://arxiv.org/pdf/2407.04811). It seems like they use vectorisation as well to achieve large speedups. More importantly, however, is that they primarily use JAX---which is becoming increasingly common in RL, and is reducing computational load significantly/making RL more accessible to compute-poor labs/institutions. Could you please comment on a few things\n\t- How does this paper's score compare to yours?\n\t- How does the walltime compare to yours?\n\t- What do you see as the benefits/disadvantages of this hardware accelerated paradigm compared to the more classic approach you are taking?\n- I know it is not usual in these types of papers but I would really appreciate a PPO comparison, both in terms of walltime and performance." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- This paper gives RL researchers a way to do pretty well in atari without expending too significant computational resources.\n- They perform ablations on their individual changes to identify what helps and what has the most effect on performance/walltime. This is quite useful.\n\nI am not giving this a lower score because I think making RL more accessible is worthwhile, and this paper takes a step towards this, and further analyses many of these independent components to see what their effect is. I am not giving a higher score because I think the paper's significance does not warrant it." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper combines several different RL improvements to a single algorithm, with a focus on high performance with reasonable computational requirements. In doing so, they find that their approach achieves a new SoTA (not including recurrent methods), while being able to be run on a desktop machine in under a day.\n\nThey analyse the factors that led to this performance in detail through several ablations.\n\nOverall, this paper makes rainbow/dqn-type methods more accessible to non-industry labs" }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- To me it is unclear if 12 hours is for all games or just 1.\n- I wonder how this fits in with the recent trend of hardware-accelerated RL (see e.g., Stoix/PureJaxRL/Gymnax and all of the hardware-accelerated environments). Does that line of work better achieve the goal of making RL more accessible? In that setting, the environment is often run entirely on the GPU, leading to hundreds or thousands of times speedups." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weaknesses.\n\n\nOverall, this work is just not good enough in its current format. I recommend that the authors fix the presentation of the results, especially adding error bars and effective aggregation using a tool like rliable. Given the significant problems with every figure and table in the main body of the paper, this work is not good enough for this venue in its current form and would require wholesale changes to fix that.\n\n[1] Deep Reinforcement Learning at the Edge of the Statistical Precipice. Agarwal et al. Neurips 2021.\n\n[2] Aitchison, Matthew, Penny Sweetser, and Marcus Hutter. \"Atari-5: Distilling the arcade learning environment down to five games.\" International Conference on Machine Learning. PMLR, 2023.\n\n[3] Sokar, Ghada, et al. \"The dormant neuron phenomenon in deep reinforcement learning.\" International Conference on Machine Learning. PMLR, 2023." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "The paper has a number of positive points:\n- The core idea of trying to achieve strong performance using Q-learning on a desktop PC has significant merit and would constitute an interesting contribution\n- The introduction of new games to evaluate on is interesting and the games chosen would make good potential benchmarks.\n- The paper is easy to follow and clearly written" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors present Beyond the Rainbow (BTR), an algorithm combining advances in Q-learning based approaches to Atari developed since Rainbow. \n\nThe authors train their agent on Atari, Procgen and 3 games which aren't well-established benchmarks in RL (Super Mario Galaxy, Mario Kart and Mortal Kombat). They run ablations on their method and demonstrate that the Impala architecture contributes the most to their method's performance. They also demonstrate that vectorization of the environment is key to the faster runtime of their algorithm." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "I slightly feel for the authors of this paper. I would like to be able to commend the paper on its empirical results, or the performance on the new environments, but the results are not presented scientifically enough for me to do that, and so I can't recommend acceptance at this venue.\n\nTo make my objections more concrete:\n- In Figure 2, the authors claim that their method outperforms the red baseline, but this is plainly not the case from the plot. The error bars so significantly overlap the red line there is no way this result is significant. \n- The authors do not aggregate their results in accordance with recommended practice [1]. Although they use the inter-quartile mean, they do not provide bootstrapped confidence intervals to estimate errors and do not seem to provide errors in their baseline results. This issue appears in Figures 1 and 2. As far as I know, the authors do not state what the error bars in Figures 1 and 2. If the plotted error bars are standard \n- While the evaluation of their method on new games is nice, I can't take any information away from this without even a semblance of a baseline. Training an RL policy on Wii games has no intrinsic scientific value -- it is only by contextualisation of a baseline that this would be a compelling result. Similarly, the authors provide no error bars in this domain.\n- Figure 4 again because of the way the results were processed provides almost no information. Atari-5 [2] provides a way to estimate the median given performance on those 5 games. But this is only after the application of a regression procedure. Without the application of this summary metric, it is just not clear what to take away from these results. This figure does not even present human normalised scores, as is standard. This Figure should therefore be replaced by a plot of the regressed median for Atari-5 with bootstrapped confidence intervals. The authors can use rliable [1] for this.\n- Again, the analysis in Section 5.2 *should* be compelling and interesting reading, but it's just not done thoroughly enough. Figure 6 is presented without error bars and so are the results in Table 2 and the IQM in Table 3. It's just not possible to believe the authors' conclusions on their work without any estimates of error. \n- Additionally, the authors use dormancy [3], but set a threshold of 0.1. Although resetting dormant neurons was shown to improve performance, neurons with a small activation are not in themselves a problem! A neuron followed by a ReLU activation that always outputs 0 is not learning, which clearly constitutes a problem, but a neuron that outputs a small value is still perfectly plastic. The dormancy results therefore also aren't a proxy for any form of plasticity. \n- The authors make multiple claims about their method being \"state-of-the-art for a desktop PC\" (or similar). These should be removed from the paper as they are just impossible to verify. Even as an expert, I do not know the hardware that every paper ran experiments on and whether it would be possible to run it on a desktop PC, and it is not a claim that can be clearly backed-up. I note that the authors did not do all their experimentation on a desktop PC, but only claim that their method can run on one effectively." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We create a high-performance Deep Reinforcement Learning algorithm capable of solving even modern games on a desktop PC." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024beyond,\ntitle={Beyond The Rainbow: High Performance Deep Reinforcement Learning On A Desktop {PC}},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0ydseYDKRi},\nnote={under review}\n}" }, "abstract": { "value": "Rainbow Deep Q-Network (DQN) demonstrated combining multiple independent enhancements could significantly boost a reinforcement learning (RL) agent’s performance. In this paper, we present \"Beyond The Rainbow'\" (BTR), a novel algorithm that integrates six improvements from across the RL literature to Rainbow DQN, establishing a new state-of-the-art for RL using a desktop PC, with a human-normalized interquartile mean (IQM) of 7.4 on Atari-60. Beyond Atari, we demonstrate BTR's capability to handle complex 3D games, successfully training agents to play Super Mario Galaxy, Mario Kart, and Mortal Kombat with minimal algorithmic changes. Designing BTR with computational efficiency in mind, agents can be trained using a desktop PC on 200 million Atari frames within 12 hours. Additionally, we conduct detailed ablation studies of each component, analyzing the performance and impact using numerous measures." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Reinforcement Learning", "Computational Efficiency", "High Performance", "Atari", "Value-Based", "DQN", "Rainbow DQN", "BeyondTheRainbow" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/c56f3eb04d73f7143c9e5a5a825865d20fac2be9.pdf" }, "presentation": null, "primary_area": { "value": "reinforcement learning" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/78ee19bcfb191885cfd57f2583128722dd9bae59.zip" }, "title": { "value": "Beyond The Rainbow: High Performance Deep Reinforcement Learning On A Desktop PC" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Suggestion for Table 1: Make it clearer that the (P) and (FT) columns correspond to specific adversarial settings\n\nIt would be interesting with some more error analysis for what breaks down when the subspace hypothesis fails, to get insights into the potential for these methods to scale to more complex settings." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "Although I am not very familiar with this field, the method for identifying the binding subspaces seems quite novel, and potentially will provide useful insights into model behavior.\n\nThe task setup, although very synthetic in nature, has the PARA and TRANS variants which make it a potentially fruitful testing ground for these kinds of questions.\n\nThe general topic of understanding mechanisms and world states inside of LLMs is both interesting and important." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies how LLMs might encode propositions stated in the context, like \"Greg is a nurse. Laura is a physicist.\", by looking at the activations associated with the Greg/nurse tokens, and trying to identify \"propositional probes\" through a \"binding subspace\" of these vectors which are aligned when the proposition holds.\n\nThey use a synthetic set with 4 domains (people, countries, foods, occupations), each with a set of non-overlapping entities (from 14 to 60 per domain). They define a somewhat heuristic \"domain identifier\" probe to pick up tokens associated with each entity, and then (main novelty) use a low-rank Hessian approach to identify these binding subspaces.\n\nThere is analysis for how effective these subspaces are in changing the binding between entities (e.g., to make Greg the physicist after the context above, when answering a question like \"Therefore, Greg's occupation is\"). The conclusion is that it \"works\" to some extent, but with caveats, especially when the context gets more complicated (going from 2 to 3 entities). In addition to testing on the synthetic contexts, there is an LLM generated variant (PARA) that turns the sequence of propositions into an actual story format, and one that translates this story into Spanish (TRANS). There is non-trivial carry over of the effect to these cases. There are also comparisons to other probing baselines.\n\nFinally, they also test on some \"adversarial\" cases: 1) Prompt injection (encourage the model to answer wrongly), 2) Backdoor attacks (model fine-tuned to do badly in Spanish), 3) Gender bias (measure amount of gender bias in output probabilities for stereotypical vs anti-stereotypical occupation assignments). In all three cases they find the propositional probes are more faithful to the underlying \"true\" propositions vs the actual model output." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The \"domain probes\" to classify tokens into domain seem quite heuristic (using the mean vector of all the entities in the domain), and it seems like there could be some evaluation to see how it works (e.g., is it always the \"obvious\" tokens, like the \"Alice\" token is the name token?).\n\nSome of the decisions in the binding space design seem quite arbitrary, like \"For tractability, we parameterize x and y so that they are shared across layers\". Maybe it would then be better to just focus on a few layers? But it's perhaps fine to leave that for future investigation.\n\nFor the Prompt Injection setting (instructing the model to \"Always answer the opposite.\"), it's hard to say what a \"correct\" output should be, in fact the prompting method should probably \"ideally\" always be \"wrong\". So saying \"prompting performs worse\" is a bit confusing, although it's still an interesting result that the probing outputs are virtually unchanged." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "My main concern and suggestions on how to alleviate it is given in the weaknesses. If the authors can present evidence that helps rule out the positional/order explanation I'm more than happy to raise my score." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "**Originality:** The paper proposes a novel method for identifying a low-dimensional subspace which appears to causally mediate binding behavior in language models. Compared to the rather indirect evidence seen in prior work (Feng & Steinhardt, 2023), the present submission directly identifies this subspace, which results in a much greater degree of manipulability and interpretability.\n\n**Quality:** The method is well-motivated (§5.2) and -- at least on the data used -- works well empirically (§6.1). The qualitative analysis (Figures 5, 7, 8 and related discussion) nicely illustrates similarity in the low-dimensional dimensional subspace, as well as the limitations of the method.\n\n**Clarity:** The paper is structured well, is clearly written and flows naturally.\n\n**Significance:** Binding is generally believed to be an essential component in the formation of internal/situation/world models. As such, any progress towards understanding if/how language models perform binding on a representational level is an important contribution." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a method for finding a low-dimensional linear subspace of an LM's activation space which, so the main claim of the paper, encodes binding information.\n\nWhile *binding* is a very broad and complex concept (see Greff et al., 2020, https://arxiv.org/abs/2012.05208 for a recent overview), in this paper binding refers to the process by which textual mentions of two or more entities and attributes that participate in a certain relation are bound together into a more abstract representation of that relation. For example, understanding the text \"Alice lives in Laos\" entails recognizing \"Alice\" and \"Laos\" as mentions of entities and then forming an abstraction of their relation that can be expressed as a proposition like LivesIn(Alice, Laos). On the representational level in a neural network this requires creating internal representations of the entities in question, and then performing some transformation on those representations that signals to subsequent layers that these two representations are \"bound together\".\n\nThe main hypothesis of the paper is that this transformation can be seen as a function that takes two entity representations x and y as input and outputs their \"binding strength\" F(x, y), i.e., a score that quantifies whether the two entities are bound or not. Assuming that F is bilinear in the entity representations x and y, the authors propose a method to estimate F via its Hessian matrix H. If the binding subspace is low-dimensional, then F and the Hessian H should be low rank, which motivates the authors to analyze the rank-k truncated SVD of H. By measuring the efficacy of interchange interventions as a function of k, the authors find that a k=50 dimensional subspace mediates binding, i.e., when manipulating activations in this subspace model output changes accordingly. For example, given the input \"Alices lives in Laos. Bob lives in Peru.\" one can make the LM say \"Bob lives in Laos\" by intervening on activations in this low-dimensional subspace.\n\nHaving developed the machinery to probe an LM for internal representations of propositions, the paper demonstrates several use cases for analyzing discrepancies between the model's internal representations and its output, finding cases in which the model appears to internally represent a proposition but generates output that is inconsistent with it." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The paper does not do enough to rule out an alternative, simpler hypothesis that could potentially explain the results. Concretely, it appears possible that, due to the highly regular nature of the data, the low-dimensional subspace claimed to encode binding information primarily encodes positional information or order information. The running example \"Alices lives in Laos. Bob lives in Peru.\" has highly regular structure with fixed positional offsets between person and country names, so it is conceivable that the proposed method actually identifies a \"position subspace\" or \"entity/attribute order subspace\" and that interchange interventions claimed to modify binding information in fact modify positional or order information. The paper takes two steps into the direction of ruling out this alternative explanation, but I do not believe that they are sufficient:\n1. Using a LM to rewrite the template-generated texts into short story-like paraphrases. My concern here is that it is unclear how much of the original regularity remains in the paraphrases and how variations in the paraphrases relate to probe performance in Table 2. Since the probe performance exact match metric on the paraphrase is much lower than on the template-based data, it is possible that the probe works best on the paraphrases that are structurally closer to the templates and drops as the paraphrases become more varied and creative. An additional analysis looking at, say, probe performance as a function of token distance between entities and attributes in the paraphrases could provide evidence for or against position being encoded in the identified low-dimensional subspace.\n2. A qualitative comparison in which position and order are varied (\"parallel\" setting in Figure 5, coreference examples in Figure 7). While encouraging, these are only a few qualitative examples of representational similarities. Here, systematic, quantitative experiments would go a long way towards ruling out alternative explanations. Data for such experiments could be relatively easily generated by varying position and order in the templates, e.g., \"Alice lives currently and has always lived Laos. Bob lives in Peru\", which varies the token distance between the bound arguments or \"Alices lives in Laos. Peru is where Bob lives.\", which swaps the order of arguments. If the authors can show that the subspace mediates binding in a similar manner, this would make a much stronger justification for calling it a \"binding subspace\"" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- I noticed that the number of values of k that were evaluated is not super high, why is that? Is Hessian-based algorithm computationally intensive? \n- Separate probes are learned for each domain, but every domain contains only one predicate, do you have any sense of how well propositional probes might generalize across predicates? Is there a good way to quantify how different the probes for each domain are?\n- Do you think anything be gleaned from the singular values of H? Do they correlate with the accuracies in Figure 4 at all?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "- A new type of probe and corresponding algorithm to construct them are presented, this method is likely to be very useful to the interpretability community. Propositional probes will allow probing LLMs for the ability to represent entities as standing in certain relations to one another. \n- The subspace identified is clearly shown to be causally implicated.\n- Probes are shown to outperform prompting in adversarial setups." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a new probing technique, called propositional probes. Such probes are functions with two arguments both of which are language model representations. When applied to entities in the probe's domain, the output of the function is a symmetric metric which is expected to be high if the corresponding tokens are bound. \nA Hessian-based algorithm is proposed to find a low-rank bilinear propositional probe. The algorithm starts with a way to query the language such that giving the correct answer depends on the ability to identify if entities are bound. In the paper's experiments, the language model is asked to repeat some relational information provided in-context (e.g. which country does entity0/entity1 live in). However, the representations of the two entities are set to their midpoint, such that the Hessian reveals how the representations would have to change in order to accurately represent their binding. After the Hessian is calculated, SVD is applied and only the top k-dimensional subspace is kept. \nTo evaluate this algorithm, 'interchange interventions' are performed where the positions in the identified subspace of two (out of three) entity representations are swapped. When the model is queried, it reports the 'wrong' entity with close to perfect accuracy for some values k. The binding strength is also visualized for some example inputs.\nFurther evaluations demonstrate that the probe match prompting performance in ordinary setting, and outperform it in adversarial settings." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Side effects of the interventions are not investigated, it would be great to evaluate how much performance is/isn't lost, as an indication of how precise the interventions are.\n- Results are limited to one model." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Why did you go from proposition to text and not the other way around: use existing text (from the wild) and generate propositions from it?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper is well-written and comes with multiple contributions. The contributions include the use of propositional probes, the definition of the hessian-based algorithm and the confirmation of two hypotheses - that propositions can be decided from internal activations and that these propositions are faithful to the input context." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes a method to extract latent world states in the form of propositional probes. They form predicate-argument-argument triples for multiple domains. They propose a method based on a Hessian-based algorithm in order to identify the binding subspace. They evaluate the propositional probes in both standard and adversarial settings. For the adversarial setting they find that the propositional probes stay more faithful to the input." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- I would have liked to see a stronger focus on the adversarial experiments, in the paper. In particular, a deeper analysis on why probes remain faithful and how backdoor attacks and prompt injection could be prevented using your method.\n- The synthetic dataset setup seems very simplistic and could have been made more true to real life use, such as by using paragraphs of existing texts and extracting propositions from them." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We develop propositional probes, which extract logical propositions describing a language model's internal world state" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024monitoring,\ntitle={Monitoring Latent World States in Language Models with Propositional Probes},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0yvZm2AjUr},\nnote={under review}\n}" }, "abstract": { "value": "Language models (LMs) are susceptible to bias, sycophancy, backdoors, and other tendencies that lead to unfaithful responses to the input context. Interpreting internal states of LMs could help monitor and correct unfaithful behavior. We hypothesize that LMs faithfully represent their input contexts in a latent world model, and we seek to extract these latent world states as logical propositions. For example, given the input context ``Greg is a nurse. Laura is a physicist.'', we aim to decode the propositions WorksAs(Greg, nurse) and WorksAs(Laura, physicist) from the model's internal activations. To do so we introduce _propositional probes_, which compositionally extract lexical concepts from token activations and bind them into propositions. Key to this is identifying a _binding subspace_ in which bound tokens have high similarity (Greg $\\leftrightarrow$ nurse) but unbound ones do not (Greg $\\not\\leftrightarrow$ physicist). Despite only being trained on linguistically simple English templates, we find that propositional probes generalize to inputs written as short stories and translated to Spanish. Moreover, in three settings where LMs respond unfaithfully to the input context---prompt injections, backdoor attacks, and gender bias--- the decoded propositions remain faithful. This suggests that LMs often encode a faithful world model but decode it unfaithfully, which motivates the search for better interpretability tools for monitoring LMs." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Interpretability", "Language models", "AI Safety" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/8fd3da46525a3a2649703d2c6d894c39c3b688d4.pdf" }, "presentation": null, "primary_area": { "value": "interpretability and explainable AI" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Monitoring Latent World States in Language Models with Propositional Probes" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "please tell me the line number where you state: what the input is to the system, ie, how many views are assumed to be input? Are surface normals also assumed to be input?\nIf you both train and test on the Objaverse dataset, then why can't you report more quantitative measures of performance? Presenting small thumbnails as the research output leaves the reader wondering if the results we're viewing are just the examples that happened to work well." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "The final images do indeed look better than the rendered comparison images, for the images shown." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies the problem of recovering 3d and some material properties of the objects studied from an image (or from multiple images, or from multiple images and their normal maps. Which of these is correct was not clear from the paper).\nExtensive experimentation was performed to optimize the terms of various loss functions in order to give high-quality visual results of the re-rendered captured shapes. Extra care was taken to ensure that physically-based rendering of the captured surfaces allowed for the captured objects to be rendered under different lighting conditions.\n\nThe paper operated within a subset of the Objaverse dataset, consisting of 35k multi-view images." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "There aren't numerical results given for the system performance. Thus, the reader is constantly questioning, \"are these results cherry picked\"?\nThe paper is written for an audience of researchers operating in the same sub-field: people reconstructing 3d, training and testing of Objaverse dataset images. I'm not in that set of researchers (and ICLR readers generally won't be) and many aspects of the paper were unclear to me, see the questions below.\nThis is an engineering paper, a paper showing how to tweak parameters\nto achieve slightly better results in a very crowded field. As I read\nthe paper, I kept asking myself, \"what do I learn from this?\" and I\nrarely came up with an answer to that question. The message is,\nextensive parameter tweaking results in slightly better performance.\nI don't feel that's a message that we need to convey to the ICLR\naudience.\n\nMy concerns with the paper: \n(1) There's no high-level story presented, no obvious set of take-aways that the reader learns.\n(2) The paper doesn't present the work in a way that's accessible to readers outside of this particular subfield.\n(3) Quantitative performance evaluations are not given, just lots of thumbnail images. This is unsatisfying, and not persuasive, since the reader wonders about bad results not being shown. If the results are indeed a random selection of the system outputs, please say so.\n(4) Generalization beyond the one dataset trained on was relegated to one figure in the appendix. Same with failure cases." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "No" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The paper is easy to read. \n- The results in all figs look good, when comparing with existing methods, especially for the geometric details." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "Building large models for translating images to 3D models is now a very hot topic. This work is a follow-up in this direction. As mentioned, three things are new: 1) incorparating render-based perceptual loss into the auto-encoder training; 2) two augmentation stratigies are proposed; 3) a texturing pipeline with reference-based attention mechanism is presented. The experiments validate the effectiveness of the proposed designs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "My concerns include:\n- It lacks training details, especially the data. What is the scale of the training 3D models? Is it same with meshLRM, meshFormer and Craftsman? If no, the comparison with them may be not fair. \n\n- It lacks quantitative anlaysis of the image-to-shape models. Currently, there are only some selected examples are shown which is not enough to support the claim of SOTA accuracy. \n\n- The paper is not well-motivated. What kind of issues does this paper aim to address? This is not clear to me. \n\n- Lack technical insights. Involving of render-based perceptual loss, the propose new augmentation strategies, attention-based texturing pipeline etc. All the claimed new things are some engineering methods. I believe these can improve the performance, however it cannot brings the community new insights." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "* Please include quantitative comparisons with the latest LRM papers (e.g., TripoSR, Stable Fast3D) on datasets like GSO or OmbiObject3D.\n\nI understand that you're training a native 3D diffusion model, which is inherently more complex and resource-intensive compared to LRMs. It would be helpful to discuss whether your results are constrained by resource limitations, and to what extent techniques like rendering augmentation were necessary to achieve your results." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "* Authors provide a comprehensive image-to-3D pipeline (notably, based on 3D diffusion), which performs on par or better than LRMs, and far better than other native 3D methods\n* The paper is well-structured, clear, and easy to follow\n* A key strength is that the authors correctly point out a common issue with 3D latent diffusion models, where the outputs often look symmetric. They visually prove that geometric alignment augmentation is a well-suited solution for this problem. This is original and significant contribution of the paper.\n* Generative rendering augmentation is shown to be an effective augmentation pipeline in practice. This idea, to my knowledge, is original." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors introduce solution to image-to-3D generation problem. Instead of predicting textured mesh, they design native 3D generation approach, with a separate PBR texturing stage.\n\n* 3D generation stage consists of training a render-enhanced point-to-shape auto-encoder; they chose triplane representation instead of previously used vector set for rendering efficiency reasons.\n* Following the autoencoder training stage, they employ a diffusion UNet on top of h-stacked triplane features, with cross-attention on input image DINOv2 features\n* For texture prediction, authors train a multiview ControlNet, applied on top of Zero123++ to predict the multiview shaded renders, with another Instruct-pix2pix decomposer to separate it into PBR materials." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "* The paper lacks any quantitative evaluation, particularly in two key areas:\n * autoencoder quality: This could be benchmarked against models like 3DShape2VecSet.\n * (biggest weakness) 3D reconstruction quality (geometry-only): Without quantifiable metrics, it's difficult to assess the quality claims. Comparisons with recent LRM papers such as TripoSR and Stable Fast3D on datasets like GSO or OmbiObject3D would be beneficial.\n\n* The proposed render-enhanced autoencoder feels more like a combination of existing methods (e.g., 3DShape2VecSet + render-based loss, a technique used in prior works like DMV3D) rather than a novel contribution.\n\n* I find the justification for ray-based regularization questionable. The paper mentions that\n> render loss alone leads to severe floaters\n\n Isn't that what the BCE loss on occupancy is meant to address? If used correctly, BCE should perform as well as, if not better than, ray-based regularization.\n\n* The generative rendering augmentation seems like a training trick to artificially boost dataset diversity. While this may improve performance, it could complicate future comparisons. I'd recommend reporting metrics without this augmentation for a clearer evaluation.\n\n* Finally, the texturing pipeline appears to be a technical application of existing ideas and seems more complementary to the paper’s core contribution.\n\nIn summary, while the paper lacks core significant novelty, it presents a well-executed combination of techniques for image-to-3D problem with native 3D diffusion models." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "What is the detailed training recipes about training dataset, GPU specs, and training time. \n\nThe paper needs to show qualitative results with unseen evaluation datasets, such as Google Scanned Objects (GSO) datasets to a stronger rationale for the improved results. Only the qualitative result of the ablation study is given.\n\nFor the geometric alignment augmentation, the reviewer wonders why it is the augmentation. In L. 261-263, the authors say that “we select one view from multi-view images as the condition and rotate the point cloud’s azimuth to align the object’s orientation with the selected image as the target”. This seems to be just training with an image corresponding to each multi-view angle for a given point cloud, but it's hard to understand why this is an augmentation. The reviewer wonders if training to correspond to multi-view images for a given 3D point cloud is not an existing method for multi-view learning, and what is added differently by augmentation.\n\nFor the multi-view PBR decomposition, are the generated images between different PBR components of the same view image, and multi-view images consistent?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper provides a thorough literature survey of previous studies, analyzes the weaknesses of these methods, and presents a concrete training model and pipeline.\n\nThe qualitative results are improved than to the previous methods, especially for the fine-grained details and well-presenting to the given input view images. The human head results in Fig. 5 is the promising result, because it is out domain of the object dataset (especially for the Objaverse). It will be better to add more human head (out-domain) results.\n\nThe paper can deal with PBR textures which is essential to practical applications." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose PBR textured mesh reconstruction method from single view image.\n\nThe model has several components, which are render-enhanced auto-encoder, image-to-shape diffusion model with augmentations, and image-conditioned PBR texture generation.\n\nCompared to the previous methods, the paper shows enhanced and fine-grained textured mesh generation results." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The figures of the pipeline and modules are hard to understand and make less intuitive because of heavy abbreviation, especially for Fig. 1, Fig. 2, and Fig. 4. \n\nIt is hard to understand the whole pipeline process of the paper’s modules. Especially for the Fig. 1, the reviewer wonders what is the connection between (a) render-enhanced auto-encoder and (b) image-to-shape diffusion model with tailored augmentation. It is nice to have a well thought out and fleshed out design, but is is hard to understand the connectivity of the entire module.\n\nFor the Fig. 4 and Fig.7, it is better to denote which is the paper’s method (ours).\n\nThe paper lacks the discussion of limitations." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Geometry comparison with commercial products? I acknowledge that lower mesh quality, attributable to the constraints of high-quality data and computational resources, is a reasonable compromise. However, I am intrigued by your rationale for deeming the alignment termed \"symmetry\" as unnecessary.\n\n2. PBR material generation from scratch V.S. PBR material decomposition from shaded images? The paper presents a promising approach involving multi-view RGB image generation and subsequent multi-view RGB-to-PBR decomposition. Concurrently, an alternative methodology exists for generating albedo, metallic, and roughness attributes from scratch, as exemplified by the HyperHuman Rodin (CLAY). I am curious about your decision to opt for PBR material decomposition over this alternative technique. Additionally, I am interested in your assessment of the strengths and weaknesses of these two techniques (discussion without qualitative results is acceptable since HyperHuman Rodin is not open-source).\n\n3. More details about reference attention? Although I am not familiar with the reference attention, I am quite positively impressed by it. Could you please provide more detail?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Generative rendering augmentation appears to hold significant promise.\n\n2. The results of geometry generation demonstrate satisfactory performance relative to available open-source non-commercial methods.\n\n3. The outcomes of PBR material generation are both compelling and credible, particularly the examples of metal objects presented in the appendix." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a novel pipeline, termed MeshGen, designed for the generation of high-quality 3D meshes with physically based rendering (PBR) textures from a single image.\nDuring the geometry generation stage, the authors first utilize a render-enhanced auto-encoder to encode 3D meshes into a compact latent space. Subsequently, an image-to-shape diffusion model is trained, incorporating geometric alignment and generative rendering augmentation to address challenges related to image-shape misalignment and the model's generalization capability.\nIn the texture generation stage, the paper establishes a reference attention-based multi-view generator, which is subsequently followed by a PBR decomposer to extract PBR components, along with a UV-space inpainter to complete the rendering of occluded areas." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Robustness of PBR material decomposition? During the texture generation stage, the method initially produces multi-view shaded images, which are subsequently decomposed into their PBR components. However, the variability in light and shadow effects within these shaded images can be substantial. I am particularly interested in the robustness of the PBR decomposer. Specifically, I am curious to know whether the decomposer can effectively manage scenarios involving more intricate lighting conditions.\n\n2. PBR material generation results on more complex metal objects? The paper has demonstrated promising PBR material generation results on various metal objects, including a teapot and a roaster. I am intrigued by the potential outcomes of PBR generation on more complex metal objects, such as a game asset axe or a detailed representation of Iron Man." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024meshgen,\ntitle={MeshGen: Generating {PBR} Textured Mesh with Render-Enhanced Auto-Encoder and Generative Data Augmentation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0zGvf2yRMQ},\nnote={under review}\n}" }, "abstract": { "value": "In this paper, we present MeshGen, an advanced image-to-3D pipeline designed to generate high-quality 3D objects with physically based rendering (PBR) textures. Existing methods struggle with issues such as poor auto-encoder performance, limited training datasets, misalignment between input images and 3D shapes, and inconsistent image-based PBR texturing. MeshGen addresses these limitations through several key innovations. First, we introduce a render-enhanced point-to-shape auto-encoder that compresses 3D shapes into a compact latent space, guided by perceptual loss. A 3D-native diffusion model is then established to directly learn the distribution of 3D shapes within this latent space. To mitigate data scarcity and image-shape misalignment, we propose geometric alignment augmentation and generative rendering augmentation, enhancing the diffusion model's controllability and generalization ability. Following shape generation, MeshGen applies a reference attention-based multi-view ControlNet for image-consistent appearance synthesis, complemented by a PBR decomposer to separate PBR channels. Extensive experiments demonstrate that MeshGen significantly enhances both shape and texture generation compared to previous methods." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "3D Generation", "Texture Generation" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/a0343c0597f627fe6785409ecaaad9d62e8b31a3.pdf" }, "presentation": null, "primary_area": { "value": "generative models" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/00ac5c75561117d3535be1def028b309f6302df6.zip" }, "title": { "value": "MeshGen: Generating PBR Textured Mesh with Render-Enhanced Auto-Encoder and Generative Data Augmentation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Does the overall generation quality decrease after the proposed method?\n2. How does the motion quality changed after the proposed method?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The method is tuning-free and does not require any further training.\n2. The results outperform baseline methods. Some of the provided visual results look good." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper targets the problem of character consistency in text-to-video generation. The authors propose a training-free method to solve this problem. They find the query in the attention encodes the information of both motion and identify, which leads to the trade-off between motion dynamics and identity consistency. The experimental model used is VideoCrafter2. To solve the trade-off problem, they propose a new query injection method. Specifically, they share features between different video clips. Then, they replace the Q (query) with those from the original generation (to maintain motion from the original generation). After that, they leverage the flow map from vanilla keyframes to guide the Q injection. Their results achieve the character consistency while keeping the original motion dynamics and text alignment. The text alignment is evaluated via user study. The overall metrics for evaluation are three aspects: motion degree, id consistency, and motion text alignment." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper is relatively hard to follow regarding the details of the method part.\n2. Lack of novelty: 1. The id presevering mechanism is built upon the SDSA. the SDSA is adopted from ConsiStory [Tewel et al. 2024] with two minor modifications: (1) The attention not attend to each other with all frames from different clips, but one single frame from each clip. (2) The mask estimation use ClipSeg, rather than estimated from the cross attention. 2. The motion preserving is leveraging TokenFlow [Geyer et al. 2023] to inject the motion based on the flow from original keyframes. Thus, the method is like a A+B combination with some minor modifications.\n3. The key insight \"self-attention query features (O) encode both motion and identity\" lack experimental results to demonstrate.\n4. The results are not perfect, e.g., inconsistent hairstyles in the 3rd row of Figure 1.\n5. The evaluation does not contain the overall video generation quality and the qualitative semantic alignment scores. \n6. Minor formate issues like inconsistent figure reference: Figure 1 and Fig. 4; And strange line break at line 244 and line 291." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. The paper only demonstrates the performance of a single character across different videos, and the reviewer is curious about how the proposed method performs with multiple characters.\n2. The prompt in the paper provides overly detailed descriptions of the character. Would a more concise description impact character consistency? For example, replace the \"Cinematic, middle-aged female athlete\" in Fig.8 with \"A woman\"." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper introduces a training-free method to ensure charactoer consistency and motion adherence in producing multi-shot video sequences.\n2. This paper presents a two-phase query injection strategy to balance encoding motion and identity.\n3. A benchmark and evalution protocol are proposed to evaluate consistency of video generation." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents Video Storyboarding, a training-free method that enhances pre-trained text-to-video models to generate multiple shots with consistent characters while maintaining high video quality and responsiveness to text prompts. By leveraging self-attention query features that capture motion and identity, the method addresses the trade-off between character consistency and video dynamics through a novel query injection strategy. Experimental results show significant improvements in character consistency and motion quality, offering insights into the video generation process and the interplay of structure and motion in diffusion models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The conducted experiments are not comprehensive, including two aspects: (a) The paper only provides several comparison samples. (b) The paper misses some important baseline methods, e.g., VSTAR[1]\n2. Although the purpose of the paper is to maintain consistency of characters across different video clips, the results are not particularly good. For example, in Fig.3, the color and style of clothes change across different video shots.\n\n[1] Li, Yumeng, William Beluch, Margret Keuper, Dan Zhang, and Anna Khoreva. \"VSTAR: Generative Temporal Nursing for Longer Dynamic Video Synthesis.\" arXiv preprint arXiv:2403.13501 (2024)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "Overall, this is a well-designed and rigorously evaluated method that represents a significant advancement in generating coherent multi-shot video sequences. The authors have done a commendable job in addressing the complex challenge of maintaining character consistency while preserving natural motion.\n\nSome questions for the authors:\n\n1. Have you explored any strategies to further improve the balance between identity preservation and motion quality? Are there other techniques beyond query injection that could be investigated?\n2. How do you envision this approach scaling to longer video sequences? What additional challenges might arise, and how could the method be adapted to handle them?\n3. The user study results showed that the ConsiS Im2Vid baseline achieved the highest set consistency among the baselines. Can you comment on the strengths of this approach and how it might be combined or compared with your Video Storyboarding method?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Framewise Subject-Driven Self-Attention to maintain consistency without compromising motion\n2. Novel two-phase query injection strategy to balance identity preservation and motion quality\n3. Adaptation of refinement feature injection to both conditional and unconditional denoising steps" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a method called \"Video Storyboarding\" to generate multi-shot videos with consistent characters across different scenes while preserving motion quality." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Approach limited to short video clips, unsure how it would scale to longer videos\n2. Balancing identity preservation and motion quality is still challenging, with potential tradeoffs" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "No Ethics Concerns" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "1. **Inconsistencies in Object Appearance and Action**: In the ablation study on query preservation (`ablation_q`), inconsistencies persist. For example, in the first video, the right hand of the Muppet character appears red, while in the third shot, it is not. Additionally, although the Muppet is intended to perform aerobics in a Sesame Street setting, it merely flicks its hand briefly, failing to convey the intended action sequence.\n\n2. **Static Object Issues in ConsiStory Component Ablation Study**: In the ablation study on ConsiStory components for video generation, the rabbit character intended to surf, train, and ride a bike appears mostly static in the first and third shots. This raises the question of whether these issues stem from limitations in the base model’s dynamic capabilities. If so, would using models with stronger dynamic performance, such as Dynamic Crafter or CogVideo, potentially improve motion consistency and address these static object limitations?\n\nIf the video dynamic problem is addressed, I am willing to increase my score." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. **Training-Free Approach for Subject Consistency Across Shots**: This work offers a training-free method for generating subjects with consistent identity across varying scenarios and shot transitions, which is valuable for practical applications where maintaining coherence in subject appearance is essential.\n\n2. **Novel Insights on Self-Attention Query Features**: The authors provide fresh insights into the role of self-attention query features, demonstrating that these features effectively capture both motion and identity.\n\n3. **Query-Preservation and Q-flow Techniques**: By preserving query features during early denoising and applying a tokenflow-inspired approach to select keyframes, the method achieves partial injection of query features to adjacent frames. Although it draws heavily from ConsisStory and TokenFlow, this approach has demonstrated effectiveness in enhancing subject consistency and motion dynamics to a certain extent." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper aims to generate multi-shot videos with consistent characters in a zero-shot manner. They claim that there is a trade-off between preserving character identity and video dynamics, thereby designing a two-phase approach, Q-preservation and Q-Flow, to balance the two respects." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. **Limited Novelty in Video Storyboarding**: The innovation of the proposed video storyboarding approach is limited. The primary method relies on frame-wise SDSA, which largely mirrors the approach used in ConsiStory. The only notable difference lies in the mask source, utilizing CLIPseg and OTSU segmentation rather than cross-attention. \n\n2. **Poor Writing and Project Organization**: The paper's writing and the project page's layout hinder comprehension, making it difficult for readers to follow the key contributions the authors intend to convey.\n\n3. **Minimal Improvement over Baseline Models**: The generated video storyboarding results appear similar to those produced by existing video generation baselines like Videocrafter2 or TokenFlow encoder, with little noticeable difference in output quality.\n\n4. **Lack of Motion Dynamics**: The method demonstrates limited motion dynamics. In most video segments, the objects remain static, and in every case, the object consistently occupies the center of the frame, resulting in rigid, uninspired visuals.\n\n5. **Overclaiming the Benchmark**: The authors’ claim of establishing a benchmark based on a dataset of only 30 videos, each containing 5 video shots, is unsubstantiated. This dataset is insufficiently sized and lacks diversity, with evaluations limited to character consistency and dynamic degree, providing a narrow view that does not comprehensively assess the model's capabilities." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "A training-free approach for generating video shots of the same characters, preserving identity and motion to prompt agreement." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024multishot,\ntitle={Multi-Shot Character Consistency for Text-to-Video Generation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0zRuk3QdiH},\nnote={under review}\n}" }, "abstract": { "value": "Text-to-video models have made significant strides in generating short video clips from textual descriptions. Yet, a significant challenge remains: generating several video shots of the same characters, preserving their identity without hurting video quality, dynamics, and responsiveness to text prompts. We present Video Storyboarding, a training-free method to enable pretrained text-to-video models to generate multiple shots with consistent characters, by sharing features between them. Our key insight is that self-attention query features (Q) encode both motion and identity. This creates a hard-to-avoid trade-off between preserving character identity and making videos dynamic, when features are shared. To address this issue, we introduce a novel query injection strategy that balances identity preservation and natural motion retention. This approach improves upon naive consistency techniques applied to videos, which often struggle to maintain this delicate equilibrium. Our experiments demonstrate significant improvements in character consistency across scenes while maintaining high-quality motion and text alignment. These results offer insights into critical stages of video generation and the interplay of structure and motion in video diffusion models." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "text to video", "subject consistency", "video personalization", "motion alignment", "feature injection", "extended attention" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/170f02322ab92d2cea34d111e32da1bbf49332aa.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/24468342d85c3fc4676090987d5d5d77386b4f11.zip" }, "title": { "value": "Multi-Shot Character Consistency for Text-to-Video Generation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Why do the lines in Fig. 5 start from different initial points? It looks like the authors use different initializations, which is unfair for comparison.\n2. What is the motivation for using Hellinger distance, not the KL divergence, for example? KL also admits closed-form for the distance between multivariate Gaussians." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Fair benchmarking of the neural PDE solvers and evaluation of their robustness to the input data is very important for understanding the current state of this field and identifying the gaps in the current SOTA methods.\n2. The proposed DeepFDM method provides more accurate predictions than competitors\n3. The benchmarking procedure is well-described and could be used in other works for evaluation of the new neural PDE solvers," }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The manuscript considers the problem of benchmarking neural PDE solvers and analysis of the robustness with respect to the diverse distributions. The authors propose DeepFDM, a benchmark method, and the procedure to generate train/test data for benchmarking with quantified shifts in distributions. The main idea of the DeepFDM method is to represent finite difference approximation of a particular type of PDEs through a convolutional neural network that parameterizes variable coefficients. Therefore, given a ground-truth input/output pair, such a model fits the target coefficients over the used grid. The second part of the work is the procedure to generate data with a controlled distribution shift that helps evaluate the trained model's robustness to input data out of the distribution where training data was generated. In experiments, DeepFDM shows better robustness to input data distribution shifts for the broad classes of equations than competitors while requiring fewer trainable parameters." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The main weakness of this work is that the authors combined two different contributions in a single study: a dataset generation procedure for benchmarking neural PDE solvers and a DeepFDM method for fitting the PDE coefficients. \n2. The idea of parameterizing the finite-difference method via CNN is not new and has already appeared in other works like the smoothing operator in the multigrid methodб https://arxiv.org/abs/2102.12071 \n3. The proposed method's scalability is not discussed or compared with competitors.\n4. The presentation of the problem statement is confusing since the authors start not from the inverse problem of coefficient reconstruction but from the solution reconstruction problem." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "1. How did you condition the neural PDE solvers on the coordinate-dependent PDE parameters? By adding the spatial coordinates to the solver inputs?\n2. How did you create the spatially varying PDE parameters? Using the same method as generating the initial condition?\n3. What did your data look like exactly? You mentioned 1D and 2D problems. Which PDE in Tab. 1 is 1D, which is 2D? How large was the dataset? How large were the spatial and temporal resolutions?\n4. What learning rate did you use? What optimizer? Did you train the models autoregressively or with 1-step errors only?\n5. How did you introduce the distribution shift? Did you increase or decrease the standard deviation of the PDE parameters? How many basis functions N did you use in the beginning? Did all of them have the same standard deviation?\n6. Why is the Hellinger distance between the parameters generating the initial conditions a good measure for the distribution shift?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "A novel way of comparing neural PDE solvers against numerical methods, which is in a sense more fair to the neural PDE solvers (if the neural PDE solvers are trained on real-word data)." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper compares the performance of an inverse design method combined with a numerical solver against neural PDE solvers. The authors assume they are given data generated from a known PDE family but with unknown, spatially dependent coefficients. In this problem setting, the paper proposes to compare neural PDE solvers against numerical simulators. Since the PDE parameters are unknown, they are estimated by minimizing the difference between the output of a differentiable numerical solver and the given data. The experiments show that the proposed model converges quicker than the considered neural PDE solvers and achieves lower errors both on in- and out-of-distribution data." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper is not very clear in the type of problem it approaches. The writing could be improved to make the definition of the problem more easy to understand.\n2. The usual setup in neural PDE solvers is that the PDE parameters are known. In this setting, neural PDE solvers have already been compared to numerical methods. The authors could better motivate their specific choice of problem definition (i.e., unknown, spatially-dependent PDE parameters).\n3. The method is only a useful baseline if the neural PDE solver is trained on real-world data. When the neural PDE solver is used as a surrogate for a numerical solver, the PDE parameters would be known (since they would have been used to generate the training data).\n4. There is no inference time evaluation. Faster inference is one of the main reasons for utilizing a neural PDE surrogate instead of a numerical method like the one considered in the paper.\n5. Many experimental and model details are missing (see questions)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "You need to compare your methods fairly with both ResNet and Unet, as well as FNO, since they represent different categories—traditional neural networks versus neural operators.\n\nPlease discuss the trade-offs between finite difference and automatic differentiation in your specific context, and to provide justification for the choice of FD and AD. There is considerable evidence that automatic differentiation outperforms FD in terms of training loss.\n\nCan you explain why DeepFDM doesn't show oscillation in Fig. 5, unlike the other methods?\nHow does the computational cost and training time of DeepFDM compare to the other approaches?\nGiven that Fig. 5 doesn't show significant improvement, can you clarify what advantages DeepFDM offers in terms of training dynamics?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper highlights a need for benchmarking in PDE solutions, pointing out that while neural networks can flexibly solve a variety of equations, there’s limited systematic comparison with established numerical methods. This motivation is well-founded, especially in scientific and engineering fields that demand rigorous performance metrics.\n\n DeepFDM seems to target both in-distribution (ID) and out-of-distribution (OOD) performance, providing a structured method for generating training and test data that reflects distribution shifts. This contribution is valuable since robust OOD performance is crucial for practical applications." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents DeepFDM, a benchmark approach for learning PDE solution operators that combines the structure of traditional PDE solvers with neural networks. By leveraging the strengths of scientific computing, DeepFDM offers interpretability and aims to enhance accuracy and generalization within a specific class of PDEs, rather than acting as a replacement for neural PDE solvers.\n\nWhile DeepFDM is designed specifically for certain types of PDEs, it shows limited generalization to other PDE classes, reducing its applicability in diverse scenarios. Additionally, the paper lacks a detailed analysis explaining why DeepFDM outperforms traditional methods, which weakens the justification of its advantages. Providing a rigorous theoretical analysis with established approaches would strengthen the work and clarify the specific benefits of DeepFDM in terms of accuracy and generalization." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "DeepFDM is presented as a benchmark method; however, its applicability is limited to a specific class of partial differential equations (PDEs). The paper does not sufficiently discuss how this restriction affects DeepFDM's generalizability, particularly in scenarios that require flexibility across various forms of PDEs, such as nonlinear PDEs and complex boundary conditions.\n\nFor instance, in the case of hyperbolic equations with shock locations, finite difference methods (FDM) may struggle to accurately capture the discontinuities inherent in these solutions. This limitation could significantly impact the performance and reliability of DeepFDM when applied to a broader range of PDE types.\n\nPlease explicitly state the objectives and justify the choice of comparison methods in the context of those objectives. More specifically,\nWhy do the authors compare DeepFDM to both neural networks like ResNet and Unet, as well as neural operators like FNO? It’s unclear whether the authors aim to solve individual instances of PDEs or to learn a solution operator." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Could you explain your definition of the benchmarking method? What makes DeepFDM a benchmarking method for PDE operator learning?\n- Do you want to focus on the inverse problem or forward problem? Could you explain how you make sure it is fair to compare with FNO, U-Net, and ResNet?\n- Could you provide a detailed description on datasets and training process?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper explores benchmarking neural operator methods and OOD performance of neural PDE operators, which is meaningful in scientific computing. \n\nFor one family of PDEs, it provides a neural network based solver with coefficients inferred." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The goal of this work is to design a benchmark method for learning PDE solution operators based on numerical PDE solvers. The authors proposed DeepFDM, which focuses on one class of PDEs and takes advantage of the structure of PDEs. DeepFDM learns the coefficients of the PDEs, and distribution shifts using the Hellinger distance are quantified. The results are compared with FNO, U-Net, and ResNet." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The motivation of choosing the known family of time dependent PDEs with periodic boundary conditions, and bounded coefficients is not clear to me. The problem setup seems very restricted, and the method is only applicable to learn from initial conditions.\n- It should be made clear that if the method is data-driven/known PDE, PDE solver/operator learning in the beginning. According to my understanding, the PDE needs to be known to use the finite differences solver. DeepFDM learns an operator from the initial condition for a specific family of PDEs to the solution at next time step, and the iteratively solve for a longer time. The problem setup should be more rigorous. \n- The literature review in Section 2.1 is not well organized or well-written. The papers of PDE discovery, PINN and operator learning are mentioned without a focus. Some claims are not correct and language is vague. For example, “Lu et al. (2019) propose the DeepONet architecture, which learns PDE solution operators. However, in this case, the PDE is fully known and the PDE residual is included in the loss.” It is not correct. There is no PDE known in vanilla data-driven DeepONet. The authors may refer to Physics-informed DeepONet. “Neural PDE operators aim to learn to solve a given PDE from data, without assuming that the form of the PDE is known.” This claim is conflicting with the above point. \n- One main issue is that it ‘s not fair to compare DeepFDM with FNO, U-Net, and ResNet, since the PDE structure is known and of course it can perform better than pure data-driven methods. This makes the results not convincing. \n- There is an existing paper on distribution shift quantification: M. Zhu, H. Zhang, A. Jiao, G. E. Karniadakis, & L. Lu. Reliable extrapolation of deep neural operators informed by physics or sparse observations. Computer Methods in Applied Mechanics and Engineering, 412, 116064, 2023.\n- Some notations are clearly defined. For example, m in the dataset, A, A* and \\hat{A}.\n- There are no metrics for coefficient fields if the author considers solving the inverse problem.\n- I don't see Appendix B in the manuscript. \n- \"In this case, the solution is generated on a higher resolution grid, and then coarsened (upsampled).\" It should be \"downsampled\"." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We introduct a benchmark scientific computing approach to PDE operator learning, and a benchmark method for OOD dataset generation for PDEs" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024deepfdm,\ntitle={Deep{FDM}: A scientific computing method for Neural Partial Differential Equation ({PDE}) operators},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0zZEbHLTwf},\nnote={under review}\n}" }, "abstract": { "value": "Solving Partial Differential Equations (PDE) has long been a critical challenge in many scientific and engineering domains. Recently, neural networks have shown great promise in solving PDEs by learning solution operators from data, offering a flexible and adaptive alternative to traditional numerical solvers. Despite these advancements, there is still a need for systematic benchmarking of neural operator methods against conventional approaches and for the development of datasets representing diverse distributions for robust evaluation.\nIn this paper, we introduce DeepFDM, a benchmark method for learning PDE solution operators based on numerical PDE solvers. \nDeepFDM leverages the structure of the PDE, in order to achieve better accuracy and generalization compared to neural solvers. It is designed as a solver for a specific class of PDEs and not as a replacement for neural solvers. Moreover, because DeepFDM learns the coefficients of the PDEs, it offers inherent interpretability. We also introduce a principled method for generating training and test data for PDE solutions, allowing for a quantifiable measure of distribution shifts. This method provides a structured approach to evaluate the out-of-distribution (OOD) performance of neural PDE operators. \nOur work sets a foundation for future comparisons of neural operator methods with traditional scientific computing approaches, providing a rigorous framework for performance benchmarking, at the level of the data and at the level of the neural solver." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Partial Differential Equations", "neural operators", "solution operators", "interpretable models", "out of distribution", "dataset shift", "physical models" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/082d598b1e71de2bf2d513cbb6b3a70cf36f6c31.pdf" }, "presentation": null, "primary_area": { "value": "applications to physical sciences (physics, chemistry, biology, etc.)" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "DeepFDM: A scientific computing method for Neural Partial Differential Equation (PDE) operators" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": { "value": "I have read and agree with the venue's withdrawal policy on behalf of myself and my co-authors." } }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "- How is $\\tilde{X}$ downsampled to obtain $\\tilde{X}'$? \n- Time-series forecasting experiments are often sensitive to hyperparameters, such as learning rates and batch sizes. How were these parameters chosen?\n- Could you provide a comparison of the actual training times for the models listed in Table 2?\n- Could you include results for cases when $G=1$ and when $G$ is large (e.g., $G=16$, $G=32$)?\n- There are many unnecessary horizontal spaces between equations and sentences (e.g., L222, L226, and L301). These should be removed." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "1. This paper is well-motivated.\n2. The proposed model, TimeCAT, outperforms prior models in several time-series forecasting benchmarks." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes TimeCAT, a transformer-based model for time-series forecasting that leverages hierarchical dependencies among fixed-length patches, dynamically grouped patches, and a global token. Specifically, TimeCAT first forms dynamic groups of input patches, then captures both fine-grained and coarse-grained information through hierarchical dependencies using Context-Aware Mixing Blocks. Experimental results demonstrate that TimeCAT outperforms previous models on several time-series forecasting benchmarks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The presentation quality of this paper is unsatisfactory.\n - The method is not clearly described throughout the paper. The descriptions are overly wordy, with many unnecessary and confusing notations.\n - For instance, from group ratios $\\mathbf{r} \\in \\mathbb{R}^G$, how are group indices obtained, and how are they optimized? Such indices are often assigned through discretization, making it unclear how they can be optimized using gradient-based methods. In particular, it is unclear how to compute gradients for $\\mathbf{W}\\_g$ and $\\mathbf{b}\\_g$. Additionally, as group size is calculated using the ceiling function, the optimization of the embedding $\\mathbf{l}\\_{s_i}$ is also unclear.\n - Other confusing notations are as follows: What are $\\mathbf{g}'\\_{n,i}$ in Eq(3) and $\\tilde{\\mathbf{g}}\\_{n,i}$ in Eq (7)? Is $RD$ a single hyperparameter or a product of two hyperparameters, $R$ and $D$?\n - Notational consistency is also an issue, with symbols such as $\\textbf{X}$ vs $X$ and $x$ vs $\\tilde{x}$ vs $\\hat{x}$. The frequent use of accents and subscripts/superscripts could easily confuse readers.\n2. Claims are not well supported.\n - The paper emphasizes the importance of the dynamic grouping mechanism. However, the experiments show that $G=2$ is sufficient to achieve good results. This very small number of groups does not adequately demonstrate the necessity of grouping for solving the problem, as all groups may still be too coarse.\n - Additionally, the grouping is determined by a single linear transformation of the input matrix, which raises doubts about the quality of the grouping.\n - Why does training become unstable without Eq (12) and (15)? Furthermore, there is no ablation study to justify the inclusion of this gradient detachment technique.\n - What is the rationale for the order of operations in the context-aware mixing block? One could simply apply self-attention across all tokens to capture intra-group and inter-group dependencies. If $G$ is as small as used in this paper (e.g., $G=2$), the computational complexity is not significantly greater than that of TimeCAT. The hierarchical design lacks both a clear explanation and experimental validation.\n - Why does Figure 5 highlight the effectiveness of the grouping strategy? For example, I cannot see any alignment between Figures 5(a) and 5(b), and the t-SNE plots in Figures 5(c)-(e) show no meaningful pattern. Why should distinct clusters of global tokens reflect effective separation of variables and high-level interactions? A more detailed explanation would be helpful." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. The writing of the article needs improvement. In Equation 3, what does g' mean?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. TimeCAT's hierarchical structure, with its focus on intra-group, inter-group, and global interactions, provides a comprehensive framework for capturing both local and global temporal patterns. This multi-scale modeling is a significant advancement in the field.\n2. The paper demonstrates a substantial reduction in computational complexity by applying self-attention within groups rather than across the entire sequence. This efficiency gain is crucial for handling longer sequences and larger datasets." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "TimeCAT is designed to address the limitations of existing Transformer-based models that struggle with capturing complex temporal dependencies and suffer from computational inefficiencies, particularly with long sequences. The core innovation of TimeCAT is its dynamic grouping mechanism, which segments input sequences into semantically coherent groups, enabling efficient hierarchical mixing at different levels of context. This approach facilitates the modeling of both local patterns within groups and global trends across the entire sequence." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The main experiments in the article are limited. It might be worth considering adding short-term experiments and incorporating new datasets. For example, there are many new datasets available here: https://huggingface.co/datasets/Salesforce/lotsa_data.\n2. How does the model’s performance change as the input length increases?\n3. Wouldn’t using the downsampled version of x to perform group division result in information loss?\n4. The spacing between the image and the title is too large." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "I have a few suggestions which may improve the quality of the paper:\n(a) Please revise Section 3 to enhance its readability.\n(b) Include a qualitative comparison to demonstrate improvements in computational complexity in terms of parameters, FLOPs, and memory.\n(c) I recommend including transformer-based baselines that aim to improve efficiency." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The proposed TimeCAT model was evaluated against benchmarks, and its accuracy shows promising results compared to recent state-of-the-art baselines. The Context-Aware Mixing Block enhances information exchange. In particular, the grouping mechanism allows the transformer to capture dependencies between highly related patches, thereby eliminating unnecessary computations between less related patches and improving efficiency. This mechanism demonstrates potential for effectively extracting information from Multivariate Time Series (MTS) data." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This manuscript introduces a framework that employs a hierarchical context-aware transformer, TimeCAT, designed to dynamically group time series patches and efficiently capture dependencies. TimeCAT was compared with recent state-of-the-art methods, and the results demonstrate promising accuracy." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The hyperparameter path size, \\(P\\), is crucial for capturing temporal dependencies and reducing computational complexity. This manuscript notes that in scenarios where \\(P\\) is significantly greater than \\(G\\), increasing \\(G\\) leads to higher computational savings. However, a detailed discussion on the size of \\(P\\) is absent. A parameter sensitivity study regarding \\(P\\), along with a table similar to Table 3 to highlight the computational savings, would strengthen the manuscript. Additionally, providing a specific example that compares the complexity reduction in terms of parameters, FLOPs, and memory would be beneficial.\n\nThe clarity of the context-aware mixing workflow could be improved by detailing the dimensions of all intermediate tensors.\n\nWhile the efficiency of TimeCAT is highlighted as a key contribution, the manuscript does not compare it with other efficient transformer-based MTS forecasting baselines. This comparison would provide a more comprehensive evaluation of TimeCAT's performance and efficiency." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@misc{\ncheng2024timecat,\ntitle={Time{CAT}: Hierarchical Context-Aware Transformer with Dynamic Grouping for Time Series Forecasting},\nauthor={Yun Cheng},\nyear={2024},\nurl={https://openreview.net/forum?id=0ziGSo4uWp}\n}" }, "abstract": { "value": "Transformer-based models have achieved significant success in time series forecasting by modeling global dependencies through self-attention mechanisms. However, these models often rely on fixed patch settings with locality constraints, tokenizing time series into spatially connected sub-series. This approach can hinder the capture of semantic relationships and lead to computational inefficiencies, especially when dealing with long sequences with complex temporal dependencies. \nIn this work, we introduce \\textbf{TimeCAT}—a \\underline{Time} series \\underline{C}ontext-\\underline{A}ware \\underline{T}ransformer that dynamically groups input sequences into semantically coherent groups, enabling efficient modeling of both local and global dependencies. By appending group and global tokens, TimeCAT facilitates fine-grained information exchange through a novel \\emph{Context-Aware Mixing Block}, which utilizes self-attention and MLP mixing operations. This hierarchical approach efficiently models long sequences by processing inputs in structured contexts, reducing computational overhead without sacrificing accuracy.\nExperiments on several challenging real-world datasets demonstrate that TimeCAT achieves consistent state-of-the-art performance, significantly improving forecasting accuracy and computational efficiency over existing methods. This advancement enhances the Transformer family with improved performance, generalization ability, and better utilization of sequence information." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": { "value": [ "~Yun_Cheng3" ] }, "authors": { "value": [ "Yun Cheng" ] }, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Time Series", "Context-Aware", "Transformer" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": { "value": "cheng|timecat_hierarchical_contextaware_transformer_with_dynamic_grouping_for_time_series_forecasting" }, "pdf": { "value": "/pdf/35ba6b859749a6330a375c222f72e8d34c0e6912.pdf" }, "presentation": null, "primary_area": { "value": "learning on time series and dynamical systems" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "TimeCAT: Hierarchical Context-Aware Transformer with Dynamic Grouping for Time Series Forecasting" }, "venue": { "value": "ICLR 2025 Conference Withdrawn Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Withdrawn_Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "- It is unclear how the plan tree is constructed. For example, is it correct to interpret that (A ? B) ? C and A ? (B ? C) are non-isomorphic plans?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- S1: The evaluation experiments used six types of workloads, including up to five-table join queries, and achieved a higher Q-error than MSCN (CIDR2019). It was confirmed that the fine-grained model, H1, achieved the fastest convergence in training and the highest accuracy.\n\n- S2: Since machine learning-based methods often underperform in high-percentile cases, the fallback mechanism is beneficial in practice. As shown in Table 4, the deeper hierarchical fallback mechanism achieves a higher Q-error.\n\n- S3: As a query-driven approach, the featurizer’s ability to characterize predicates is useful." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a workload-driven approach for cardinality estimation aiming at workloads containig repetative queries. The proposal utilizes multiple templatizer to derive hierarchical cardinality estimation models, taking a query plan tree as input and obtaining data with different granularity. It employs general predictors like PostgreSQL to perform cardinality estimation at the lowest level." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- W1: In addition to Q-error, P-error should be also evaluated, which is becoming a standard metric.\n\n- W2: As shown in Table 3, H1 demonstrates the highest performance, so a structure of H1 -> PostgreSQL seems optimal, making the multiple hierarchy setup (in Equation 15) unnecessary. The authors should clarify the benefit of using multiple hierarchy levels. \n\n- W3: Although the proposal adopts a workload-driven approach, recent trends favor data-driven or hybrid approaches. Notably, data-driven approaches have the advantage of robustness for unknown queries. Combining a workload-driven approach with data-driven methods could enhance accuracy in cases where prediction errors are large. While a workload-driven approach has the advantage of faster inference time, it is not obvious workload-driven approach alone is useful. \n\n- W4: In Section 2.5, the statement \"All graphs whose templates are isomorphic share the same model\" is based on graph isomorphism as defined in Definition 1, relying on edge information only. The templatizer, H1, thus does not use graph attribute information. However, Section 3.1 states, \"Hence, query graphs found in the same H1 template differ only by the constant values,\" which appears contradictory.\n\n- W5: Section 3.3 includes a calculation for the hash size, such as s(#T1), but according to Equation 6, the hash length is fixed at h, so the case distinction in Equation 10 does not seem valid.\n\n- W6: Although Cardbench is used as the evaluation benchmark, it is proposed in an arXiv paper and is not yet a standard benchmark. It would be better to use the widely accepted join order benchmark or explain the strong justification for using Cardbench." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "Please refer to W1-W4." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "S1. The authors tackle an important issue in analytical databases: cardinality estimation for repetitive queries.\n\nS2. The authors propose a cardinality estimator, which leverages hierarchical localized models that are trained online." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose a supervised cardinality estimation method that enhances accuracy for workloads with repetitive queries (i.e., queries repeated in templates with only the constant parameters changed) by leveraging hierarchical, online localized models. This approach transforms each SQL query into a Directed Acyclic Graph (DAG) representation, groups isomorphic DAGs into the same partition, and trains an online model for each partition as queries are executed. Grouping is conducted hierarchically at multiple levels of granularity, ranging from fine-grained (i.e., queries varying only in constant terms are placed in the same partition) to coarse-grained (e.g., queries varying only in constants, operator types, and column names are placed in the same partition). During runtime, given a query, the method begins with the fine-grained model and moves to coarser-grained models until it finds a confident model for the query. If no suitable model exists, it defaults to a learned or traditional model. Using the CardBench benchmark, the authors demonstrate that this method yields more accurate cardinality estimates compared to competitors such as PostgreSQL and MSCN." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "W1. The Experiments section needs to be improved.\n- The authors should compare their method with state-of-the-art (SOTA) data-driven cardinality estimation methods such as [1]. Currently, the comparison is limited to MSCN, a query-driven method proposed in 2019 [2], and PostgreSQL, a traditional estimator.\n- The “Imperfect Admission Experiments” in Section 4 seems unfair, as the q-error percentiles of each method are reported on different query subsets: PostgreSQL is evaluated on the entire query set, MSCN on a subset excluding disjunctions, while the proposed method seems to be evaluated on a subset of simple queries (e.g., repetitive queries varying only in constant terms).\n- The authors should report end-to-end time (including both planning and query execution time), as shown in [1,3], along with its breakdown for further clarity.\n- Reporting training time and model size would provide further insights into the method’s practical feasibility.\n- The authors should clarify why the q-error is higher in certain intervals with larger training sample sizes in Figure 4.\n- The experimental setup requires a more thorough explanation, particularly regarding how query repetitiveness was generated and its extent.\n\n\nW2. Motivation is inadequate.\n- The authors should explain why existing learned estimators struggle with repetitive workloads to highlight the necessity for their proposed method.\n- The authors should clarify why a hierarchical model structure is effective in improving the accuracy of cardinality estimation.\n\nW3. The presentation needs to be improved.\n- The process of generating a DAG from a query needs further explanation. If the DAG refers to a query plan, the authors should specify which query plan is used, as multiple plans can exist for a single query.\n- There are some undefined terms, such as d_{\\psi} in Section 2.\n- There are inconsistent notations throughout the paper. For instance, “query graph” and “query plan” are used interchangeably.\n- Numerous typos appear throughout the paper, such as “geoping” and “hases” in Section 5.\n\nW4. There are some misstatements regarding existing work\n- The authors seem to overstate the limitations of existing methods. They claim that “NN-based estimators perform well if they are trained with large amounts of query samples,” which is true specifically for query-driven learned estimators, not all NN-based methods.\n- The authors state that “50% of the real world clusters have more than 90% queries repeated in templates (only changing the constant parameters),” citing [4]. However, according to [4], the correct value is 80%, not 90%.\n\n[1] Kim, Kyoungmin, et al. \"Asm: Harmonizing autoregressive model, sampling, and multi-dimensional statistics merging for cardinality estimation.\" Proceedings of the ACM on Management of Data 2.1 (2024): 1-27.\n[2] Kipf, Andreas, et al. “Learned cardinalities: Estimating correlated joins with deep learning.” In Biennial Conference on Innovative Data Systems Research, 2019.\n[3] Wu, Ziniu, et al. \"FactorJoin: a new cardinality estimation framework for join queries.\" Proceedings of the ACM on Management of Data 1.1 (2023): 1-27.\n[4] van Renen, Alexander, et al. \"Why tpc is not enough: An analysis of the amazon redshift fleet.\" Proceedings of the VLDB Endowment 17.11 (2024): 3694-3706." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "NA" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Q1. The addition of mean absolute error (MAE) and relative prediction error (RPE) would allow for a more comprehensive evaluation of the accuracy of different cardinality estimators.\nQ2. It should include a broader variety and greater number of baseline methods by introducing more advanced cardinality estimation approaches. Adding comparisons with data-driven cardinality estimation methods or experiments against paradigmatic methods like QueryFormer would enhance the analysis.\nQ3. It should include comparative experiments under different workloads. Additionally, experiments on cardinality estimation with lower query redundancy should be added to provide a more comprehensive evaluation.\nQ4. Could you clarify what specific models are used at each level of cardinality estimation in this paper? This detail is not adequately explained in the manuscript." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "S1. Hierarchical cardinality estimation methods have significant advantages for cardinality estimation methods in the presence of a large number of repeated queries. They allow for the training of separate cardinality estimators for each query type, thereby saving training costs and improving the accuracy of cardinality estimation.\nS2. The method presented in this paper demonstrates strong cardinality estimation performance, and it also exhibits good convergence stability through hierarchical training.\nS3. Compared to traditional cardinality estimators, the method proposed in this paper achieves faster convergence speed with lower overhead, making it suitable for practical industrial applications." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "To address the issue of repeated queries in cardinality estimation, this paper proposes an on-line cardinality estimation method. Unlike traditional cardinality estimation approaches that rely on a single model, this study introduces a hierarchical cardinality estimation framework. Queries are categorized into three levels, with different structural classifications applied at each level. For each level, distinct estimator models are trained based on various classification templates, and evaluations are conducted hierarchically. Additionally, these models utilize an extension of Merkle-Trees to hash directed acyclic graph (DAG) query plans. Finally, an ensemble learning method is used to statistically aggregate the results and produce the final cardinality estimates. Compared to traditional cardinality estimators and the query-based cost estimation method MSCN, this approach achieves superior results." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "W1. This paper employs only the Q-Error as an evaluation metric, which can assess the stability of the cardinality estimator but does not provide an intuitive measure of its accuracy. The addition of mean absolute error (MAE) and relative prediction error (RPE) would allow for a more comprehensive evaluation of the accuracy of different cardinality estimators.\nW2. This paper compares only with two relatively outdated query-based cardinality estimation methods, MSCN and MSCN+. It should include a broader variety and greater number of baseline methods by introducing more advanced cardinality estimation approaches. Adding comparisons with data-driven cardinality estimation methods or experiments against paradigmatic methods like QueryFormer would enhance the analysis.\nW3. The experimental workload in this paper lacks clarification regarding query redundancy. It should include comparative experiments under different workloads. Additionally, experiments on cardinality estimation with lower query redundancy should be added to provide a more comprehensive evaluation." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "Given many labeled graphs, each representing database query, where graph label is cardinality, we group graphs by graph-structure and learn simple model per group (within group, feature dimension is constant)." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024hierarchical,\ntitle={Hierarchical Graph Learners for Cardinality Estimation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=0zmHFyZwkA},\nnote={under review}\n}" }, "abstract": { "value": "Cardinality estimation -- the task of estimating the number of records that a database query will return -- is core to performance optimization in modern database systems. Traditional optimizers used in commercial systems use heuristics that can lead to large errors. Recently, neural network based models have been proposed that outperform the traditional optimizers. These neural network based estimators perform well if they are trained with large amounts of query samples. In this work, we observe that data warehouse workloads contain highly repetitive queries, and propose a hierarchy of localized on-line models to target these repetitive queries. At the core, these models use an extension of Merkle-Trees to hash query plans which are directed acyclic graphs. The hash values can divisively partition a large set of graphs into many sets, each containing few (whole) graphs. We learn an online model for each partition of the hierarchy. No upfront training is needed; on-line models learn as the queries are executed. When a new query comes, we check the partitions it is hashed to and if no such local model was sufficiently confident along the hierarchy, we fall-back onto a default model at the root. Our experimental results show that not only our hierarchical on-line models perform better than the traditional optimizers, they also outperform neural models, with robust errors rates at the tail." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Cardinality Estimation", "Many small models", "Graph Hash", "Group-by-template", "Fast Learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/a797ce27036a1f3593a57a960711d98c47e24fbc.pdf" }, "presentation": null, "primary_area": { "value": "learning on graphs and other geometries & topologies" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Hierarchical Graph Learners for Cardinality Estimation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Can you justify why optimality gap and computational efficiency, which depends on solver, are also included in the evalutation metric?\n2. What is the relationship of Types I to Types III problems with the rest of this paper?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The approach to utilize LLM both as hypothesis generators and evaluators seems novel. \n2. A pruning technique based on Satisfiability Modulo Theories (SMT) solvers is introduced to eliminate redundant, trivially equivalent formulations, which can improve the search efficiency in my understanding." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper is along the popular direction of using LLM to automatically fomulate mathematical optimization problems. It introduces a formal definition of autoformulation and identifies three main challenages. To address these challenges, the authors propose a method using large language models (LLMs) within a Monte-Carlo Tree Search (MCTS) framework. LLMs are utilized both as hypothesis generators and evaluators, while MCTS incrementally explores the formulation space. Additionally, a pruning technique based on Satisfiability Modulo Theories (SMT) solvers is introduced to eliminate redundant, trivially equivalent formulations. The method was tested on benchmarks (NL4OPT and IndustryOR) and show superior results." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The concept of autoformulation defined in this paper is broad, including mathematical formulation and computational formulation, and optimality gap and computational efficiency. However, the main contribution of this paper is on mathematical formulation if I understand correctly. The rest in the proposed concept is not addressed in this paper.\n2. I do not see why adding optimality gap and computational efficiency to the evaluation metric. They're solver dependent and is not the responsbility of formulation.\n3. The authors defined Types I to Types III problems, but I do not see how they are related to the main contribution of this paper.\nOverall, I feel like the concept proposed in this paper is too broad and distract readers from the main contribution of this paper." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. In Section 5.2, paragraph (2) ‘Getting Local Scores,’ how does the DFS expand the tree? Does it expand the child with the highest prior score first, or does it expand a random model generated by the LLM?\n\n2. How should Figure 5 be interpreted? Section 5.4 doesn’t seem directly related.\n\n3. How is exploration of the hypothesis space ensured? Figure 4 shows that pruning is effective, but does this imply that the LLM fails to generate diverse partial formulations?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Leveraging LLMs for both hypothesis generation and evaluation enables verification of partial problems.\n2. Using MCTS to efficiently navigate the large hypothesis space, along with introducing the SMT pruning procedure, saves computational resources and focuses on unique, meaningful formulations.\n3. Strong experimental results compared to baselines." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces a framework to automate the creation of mathematical optimization models from natural language descriptions. This process is essential in operations research but traditionally requires significant expertise. There are three main challenges for autoformulation: defining a vast hypothesis space, navigating this space efficiently, and ensuring formulation correctness. To address these, the authors integrate LLMs within an MCTS framework, using LLMs to generate potential formulations and evaluate correctness. They also apply a pruning technique with SMT solvers to eliminate redundant formulations. Experiments on benchmarks for linear and mixed-integer programming show this approach is both accurate and efficient, potentially making optimization modeling more accessible." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The same LLM is used for both hypothesis generation and verification, which introduces a high correlation between generation and evaluation stages. This could limit the objectivity and diversity of the verification.\n\n2. Lacks a clear explanation of how the LLM is guided to explore the hypothesis space with sufficient diversity.\n\n3. It would be beneficial to provide more insight into partial problem formulation at intermediate steps. Improved verification of partial models could enhance accuracy and reduce runtime for complex problems.\n\n\nMinor issue: The font size in the figures is too small" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Could you clarify how the developed system will be utilized by domain scientists or in other potential use cases?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The paper provides a clear and compelling description of the motivation behind the research, along with a well-articulated explanation of the proposed method.\n- It includes an extensive experimental comparison that effectively demonstrates the performance and advantages of the proposed approach relative to existing methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper identifies three core challenges: defining the vast, problem-dependent hypothesis space, efficiently searching this space under uncertainty, and evaluating the correctness of formulations. \n\nTo tackle these issues, the authors propose a novel method that leverages Large Language Models (LLMs) within a Monte-Carlo Tree Search framework. LLMs function as both hypothesis generators and evaluators of formulation correctness, while a pruning technique is employed to eliminate trivial equivalent formulations. \n\nEmpirical evaluations demonstrate that this approach significantly improves the efficiency and accuracy of formulating optimization models for linear and mixed-integer programming problems, making it accessible to users without deep optimization expertise." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- In my experience with Operations Research, many papers focus on formulating new problems into linear programming or mixed-integer linear programming. These papers typically begin with a clear textual definition of the problem, followed by rigorous proofs, given the critical applications involved. However, I find that this work may not be particularly useful for practitioners in the field, as they still need to undertake similar formulation efforts themselves. The primary benefit seems to lie in providing students with examples for their coursework rather than advancing practical applications for researchers and professionals.\n\n- In line 128, the quote from the textbook stating, \"Once this formulation is done, solving the problem is ... (almost) technology,\" highlights a distinction from the problem formulation presented in this paper. The textbook emphasizes identifying the context of the problem, extracting the decision variables, and compiling all relevant real-world constraints, which cannot be adequately captured through mere textual descriptions." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Questions:\n1. Why having a probabilistic model $p_\\psi$ with complex joint optimization in Section 2.1? (See Weaknesses 1 for details)\n\nSuggestions:\n1. Highlight the relation and difference between this paper and other works combining LLMs with MCTS.\n2. Align the problem definition and the methods. If the quality of computational representation is not sufficiently addressed in this paper, the problem definition may not be so complex. (In my opinion, a method which fit the current problem definition will be, for example, bi-level finetuning of two LLMs $p_\\phi$ and $p_\\psi$ to maximize $Q$)\n3. It may worth fine-tuning the model with MCTS-based self enhancement in the future (check arXiv:2309.17179 for an example)." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Good writing, easy to follow.\n2. Specific improvements on MCTS to fit the optimization scenario (merging trivially equivalent formulations, including solver signals to rewards, etc.)\n3. Sufficient experimental details." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposed a MCTS-enhanced LLM approach, to transform descriptions of certain types of optimization problems into formal optimization models consisting of (1) variables, (2) objective, (3) equality constraints and (4) inequality constraints. The MCTS has a depth of 4 corresponding to the four components of an optimization model. When expanding a node, LLMs are used to generate a set of candidate formulations, and trivally identical formulations are pruned by SMT solvers to improve efficiency. The expanded nodes will be assigned a prior by an LLM-based ranking evaluation. The reward at the terminal node is a combination of LLM evaluation and whether optimality is reached from solvers. Experimental results show that the proposed approach outperforms other non-MCTS approaches." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The rationality of the definition of \"autoformulation\" is not sufficiently backed by the main part of the paper. The connection between the problem setting and the proposed method is not very clear. For the definition, two transformations are highlighted (mathematical formulation model $p_\\phi$ and computational representation model $p_\\psi$), and \"autoformulation\" is defined to jointly optimize $(\\phi, \\psi)$ so that the expection of a quality function $Q$ is maximized. This problem setting is a bit different from other \"LLM+OR\" papers. However, in the main part of the paper, no clear effort is paid on optimizing parameters, neither $\\phi$ nor $\\psi$ (GPT-4 is used without fine-tuning). I feel that the main objective of this paper is still the same as other papers, proposing a $P_{\\phi\\text{-LLM}}^\\text{improved}$ that empirically performs better than a vanilla use of $P_{\\phi\\text{-LLM}}$. So I am a bit confused why bother to propose such a definition. Specifically, it seems unneeded to have a probabilistic model $p_\\psi$ with complex joint optimization, if the computational representation step can be simply done by a deterministic parser (line 241) or commertial packages (line 194).\n2. There are already lots of literatures working on combining LLMs with MCTS to solve general or domain-specific problems (see arXiv:2309.17179, arXiv:2402.03289, arXiv:2409.09584, arXiv:2406.07394, arXiv:2402.08147 for some examples), in both training and inference stage, which may unfortunately raise the bar for another \"LLM+MCTS\" paper to appear on top conferences. It is good to have LLM+MCTS applied on OR problems, but more unique features might be required to distinguish this paper. This paper has paid some effort on it (see Strength 2), but apart from these adaptations, the whole LLM+MCTS pipeline seems quite standard, and only applied at the inference stage.\n3. While LLM contains general knowledge of optimization, the proposed method is limited to (reformulated) convex problems (type I and II in this paper)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024autoformulation,\ntitle={Autoformulation of Mathematical Optimization Models Using {LLM}s},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=107ZsHD8h7},\nnote={under review}\n}" }, "abstract": { "value": "Mathematical optimization is fundamental to decision-making across diverse domains, from operations research to healthcare. Yet, translating real-world problems into optimization models remains a formidable challenge, often demanding specialized expertise. This paper formally introduces the concept of *autoformulation*---an automated approach to creating optimization models from natural language descriptions for commercial solvers.\nWe identify the three core challenges of autoformulation: (1) defining the vast, problem-dependent hypothesis space, (2) efficiently searching this space under uncertainty, and (3) evaluating formulation correctness (ensuring a formulation accurately represents the problem).\nTo address these challenges, we introduce a novel method leveraging *Large Language Models* (LLMs) within a *Monte-Carlo Tree Search* framework. This approach systematically explores the space of possible formulations by exploiting the hierarchical nature of optimization modeling. LLMs serve two key roles: as dynamic formulation hypothesis generators and as evaluators of formulation correctness. To enhance search efficiency, we introduce a pruning technique to remove trivially equivalent formulations. \nEmpirical evaluations across benchmarks containing linear and mixed-integer programming problems demonstrate our method's superior performance. Additionally, we observe significant efficiency gains from employing LLMs for correctness evaluation and from our pruning techniques." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Large Language Models", "optimization modeling" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/6da87802a6b8d8f636375a46bc11a1eecc11f30d.pdf" }, "presentation": null, "primary_area": { "value": "applications to robotics, autonomy, planning" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Autoformulation of Mathematical Optimization Models Using LLMs" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "- Defining the logic rule set seems expensive. Would it be difficult to construct a factor graph by connecting concepts and categories with a bipartite complete graph and estimating the weights $w_i$?\n- How do you distinguish between coexistence and exclusion in the factor graph?\n- It would be better to describe the specific estimation algorithm of $w_i$ in the Appendix.\n- Line 336-337: What is \"attributional training\"?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- The proposed method enables both the detection of logical errors and the correction of these errors within a single framework.\n- Compared to other concept-based interpretable neural networks, the proposed method achieves higher comprehensiveness in explanations, regardless of whether the perturbations are known or unknown." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposed a method using factor graphs to correct errors in concept-based explanations caused by perturbations. \nSpecifically, it constructs a factor graph using predefined logical rules between concepts or between concepts and categories. \nThis graph helps identify logical errors in the output from concept-based interpretable neural networks by evaluating the likelihood of these errors. \nAdditionally, by leveraging the factor graph, it is possible to correct these logical errors in the explanations. \nExperimental comparisons on three datasets demonstrate that the proposed method outperforms existing concept-based approaches." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- While the proposed method assumes that explanations change due to perturbations without affecting predictions, this seems unrealistic. Particularly in interpretable neural networks with a concept bottleneck structure, as assumed in this study, changes in the concepts outputted by the neural network would naturally lead to changes in predictions, which undermines this assumption.\n- The proposed method requires predefined logic rules between concepts and categories. If these rules are already known, wouldn’t it be possible to detect inconsistencies between concepts and predictions without the need for the factor graph? The advantage of using a factor graph is unclear.\n- As noted in the minor comments below, there is room for improvement in the writing.\n\nMinor comments:\n- The explanation of Figure 3 is insufficient.\n- There is no reference to Figure 4." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Can the authors provide a comparison against one of the methods injecting knowledge (prior or learnt)?\n- Scalability: The two limitations that have been reported (domain knowledge changes, correct prediction categories) are non-negligible. How could this method be extended to face these limitations?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- **Good presentation and structure** The paper is well structured and easy to read. The mathematical definition of the proposed method is clear and well-defined. \n- **Nice experimental campaign**: Extensive experiments on three datasets (CUB, MIMIC-III EWS, and Synthetic-MNIST) demonstrate the superior performance of AGAIN over the compared baselines." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces AGAIN (fActor GrAph-based Interpretable Neural Network), which generates comprehensible explanations for neural network decisions even under unknown perturbations. Unlike traditional adversarial training methods, AGAIN incorporates logical rules through a factor graph to identify and rectify explanatory logical errors during inference. The model is validated on three datasets: CUB, MIMIC-III EWS, and Synthetic-MNIST, demonstrating superior performance compared to existing baselines" }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "## Major Issues\n\n- **Novelty, Related work and Compared methods**: the main issue with the current paper is that it only considers methods injecting knowledge into the models by means of factor graphs. However, the integration of knowledge into model predictions has been fully studied under different point of views: probabilistic (e.g., DeepProblog [1] and all its variants), logic constraints (see the survey of Giunchiglia et al. [2]). Also, it is not the first method defending against adversarial attack with knowledge and without employing adversarial training. Some of these methods have been already employed to defend against adversarial attacks, such as [3-4]. [5] is a survey entirely dedicated to enhancing interpretability and adversarial robustness with prior knowledge. [6] has shown in the context of concept-based models that it can learn logic rules at training time and use them at inference time to defend against adversarial attacks. This is also reflected in the experimental campaign that is extensive but does not consider any methods injecting prior knowledge to defend against adversarial attacks. The only compared methods are CBMs or adversarial-trained versions of the same models. \n\n\n- **Paper positioning and Preliminaries**: the method provides explanations and a defence mechanism that is based on concept predictions; thus, it applies only to concept-based models. Most of the compared methods also belongs to this niche. While this is fully acceptable, explicit references to concept-based models only appears in Section 3-4. Therefore, I think it should state earlier that this is the focus of the paper, as most of the related work mentioned in the paper does not focus on concept-based explanations. Furthermore, there is no explicit mentions to concept-based models in the preliminaries. The “Interpretable Neural Network” paragraph should include citations to this literature and explain concept-based models.\n\n## Minor Issues\n- P.2 “[…] even if the adversarial samples are available, retraining is only effective for at most one perturbation type” I think this statement is quite strong, and although it may have been proved for some attacks in Tan & Tian, 2023, I don’t think it is a general assumption that we can make. I think this sentence should be soften to “retraining is effective only for few perturbation types at the same time”. \n- P.2 “[…] to ensure the expectation of the potential function is in its maximum value”. It is not clear at this point what is the potential function. Consider rephrasing this sentence.\n- P.2 “The explanations that are further regenerated align with the exogenous logic.” Not clear in this case what are the exogenous factors. \n- P.2-3: The term \"defenses against comprehensibility\" seems a bit misleading. It implies that the goal is to prevent explanations from being understandable, which is not the case. Instead, the focus is on defending the comprehensibility of explanations against perturbations. “defenses of comprehensibility” colud be a more appropriate definition.\n\n\n[1] Robin Manhaeve, Sebastijan Dumancic, Angelika Kimmig, Thomas Demeester, Luc De Raedt: DeepProbLog: Neural Probabilistic Logic Programming. NeurIPS 2018: 3753-3763\n\n[2] Giunchiglia, E., Stoian, M. C., & Lukasiewicz, T. (7 2022). Deep Learning with Logical Constraints. In L. D. Raedt (Ed.), Proceedings of the Thirty-First International Joint Conference on Artificial Intelligence, IJCAI-22 (pp. 5478–5485). doi:10.24963/ijcai.2022/767\n\n[3] Yin, M., Li, S., Cai, Z., Song, C., Asif, M. S., Roy-Chowdhury, A. K., & Krishnamurthy, S. V. (2021). Exploiting multi-object relationships for detecting adversarial attacks in complex scenes. In proceedings of the IEEE/CVF international conference on computer vision (pp. 7858-7867).\n\n[4] Melacci, Stefano, et al. \"Domain knowledge alleviates adversarial attacks in multi-label classifiers.\" IEEE Transactions on Pattern Analysis and Machine Intelligence 44.12 (2021): 9944-9959.\n\n[5] Mumuni, Fuseini, and Alhassan Mumuni. \"Improving deep learning with prior knowledge and cognitive models: A survey on enhancing interpretability, adversarial robustness and zero-shot learning.\" Cognitive Systems Research (2023): 101188.\n\n[6] Ciravegna, Gabriele, et al. \"Logic explained networks.\" Artificial Intelligence 314 (2023): 103822." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "How is the reasoning conducted on G? How is the probability estimation implemented in Equation 1? Please provide more details." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper proposes a novel and interesting idea that uses the logical correctness of explanation to detect and defense against noises and adversaries.\n2. The three stage framework is reasonable to me. \n3. The examples used in the paper are intuitive." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes AGAIN (fActor GrAph-based Interpretable Neural Network), a new approach to maintain comprehensible neural network explanations when inputs are affected by unknown perturbations. AGAIN builds factor graphs from explanations, and integrates logical rules through the graphs. The system detects adversaries by identifying violations of real-world logic, and uses an interactive intervention strategy to fix incorrect explanations. Tested on three datasets (CUB, MIMIC-III EWS, and Synthetic-MNIST), AGAIN outperformed existing methods in generating comprehensible explanations under both known and unknown perturbations." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The concept bottleneck model is not easy to scale, as it requires manual annotations of concepts.\n2. The implementation details of Section 4.2 is not very clear (the introduction is too conceptual). For example, is $\\mathcal{G}$ a graph or a model? At the beginning I thought $\\mathcal{G}$ is a graph, but in Line 221 it \"reasons about a conditional probability\".\n3. The datasets used in this work are not very strong. I doubt if the work is applicable to real-world situations. At least, the datasets used cannot reflect adversarial scenarios in practice." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. The authors employed black-box perturbations to test robustness. Adding input noise could provide more convincing evidence.\n2. I am concerned about the potential impact of large factor graphs on computational efficiency. I would like to see an analysis on this problem.\n3. Currently, the algorithm explores all intervention options to select the optimal one. Is it possible to employ a simplified strategy, such as a heuristic or greedy algorithm, to reduce computational load?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper presents an innovative method, AGAIN, that combines factor graphs with concept-level explanations to improve model interpretability under unknown perturbations.\n2. The authors evaluate AGAIN across multiple datasets and baseline models, providing a broad view of its effectiveness.\n3. The paper is clear and well-structured." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes AGAIN, a neural network model that generates comprehensible explanations under unknown perturbations by integrating logical rules directly during inference, rather than relying on adversarial training. Using a factor graph to identify and rectify logical inconsistencies, AGAIN demonstrates superior interpretability and robustness compared to existing methods." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The factor graph requires predefined logical rules, which could be challenging to construct or generalize across different domains.\n2. The algorithm’s process of exploring all possible intervention options to find the optimal solution could create computational overhead." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See Weaknesses" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper offers a well-motivated and clear presentation of a unique approach. The use of a factor graph to handle perturbations is innovative and highly relevant to the research community. Furthermore, the authors have provided sufficient theoretical support and empirical evidence to justify the effectiveness of their approach." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the authors propose a novel method called AGAIN to generate comprehensible explanations under unknown perturbations by employing a factor graph approach. The method presents a significant contribution to the field, addressing an important gap with a new perspective. The paper is well-structured, and the authors have provided thorough theoretical justifications and experimental analyses. These analyses effectively demonstrate the superiority of the proposed method over existing ones." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Clarification on Figure 3:\n In Figure 3, does the final output include hat{c}_{re} and prediction? Does the hat{c}_{re} mean the interpretation? Additionally, the figure lacks clarity on how hc is generated, which is crucial for understanding the method. Adding details on hc generation would make the figure more comprehensive.\n \n2. Comparison Metrics:\n The authors compare their method with all baselines using the LSM metric. However, they only compare their method with CBM in terms of accuracy (P-ACC and E-ACC). It would be beneficial to extend the accuracy comparison to include all baselines to provide a more complete evaluation of the method’s performance.\n \n3. Figure 7 Interpretation and Readability:\n In Figure 7, for each example, the authors provide two bar charts. Does the left bar chart represent the initial interpretation, and the right bar chart represent the interpretation combined with the factor graph? Clarification on this aspect would enhance the understanding of the figure. Additionally, some symbols and text in Figure 7 are overlapping.\n \n4. Inconsistency in Line 463 and Table 5:\n The authors mention in line 463 that their method is compared with two baselines on the Synthetic-MNIST dataset. However, Table 5 lists four baselines. Furthermore, \"ProbCBM\" in Table 5 should be corrected to \"ProCBM\" for consistency. It is recommended that the authors proofread the paper to eliminate such inconsistencies and typographical errors." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024factor,\ntitle={Factor Graph-based Interpretable Neural Networks},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=10DtLPsdro},\nnote={under review}\n}" }, "abstract": { "value": "Comprehensible neural network explanations are foundations for a better understanding of decisions, especially when the input data are infused with malicious perturbations. Existing solutions generally mitigate the impact of perturbations through adversarial training, yet they fail to generate comprehensible explanations under unknown perturbations. To address this challenge, we propose AGAIN, a fActor GrAph-based Interpretable neural Network, which is capable of generating comprehensible explanations under unknown perturbations. Instead of retraining like previous solutions, the proposed AGAIN directly integrates logical rules by which logical errors in explanations are identified and rectified during inference. Specifically, we construct the factor graph to express logical rules between explanations and categories. By treating logical rules as exogenous knowledge, AGAIN can identify incomprehensible explanations that violate real-world logic. Furthermore, we propose an interactive intervention switch strategy rectifying explanations based on the logical guidance from the factor graph without learning perturbations, which overcomes the inherent limitation of adversarial training-based methods in defending only against known perturbations. Additionally, we theoretically demonstrate the effectiveness of employing factor graph by proving that the comprehensibility of explanations is strongly correlated with factor graph. Extensive experiments are conducted on three datasets and experimental results illustrate the superior performance of AGAIN compared to state-of-the-art baselines." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "interpretable neural network", "factor graph", "perturbation", "explanation rectification", "graph learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/e27cefe47e843cdc9ae47f185ccbf3238aac30a7.pdf" }, "presentation": null, "primary_area": { "value": "interpretability and explainable AI" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Factor Graph-based Interpretable Neural Networks" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "The waveform template is restricted to one channel with maximal amplitude and multi-channel template results are shown in Supplement E. What exactly is the definition of a channel in this context?\n\nWhy was additive Gaussian noise chosen as the sole data augmentation? Can a brief rationale for this specific choice be included in the paper? The authors demonstrate that adding two template augmentations: amplitude jitter and electrode dropout does not significantly improve the performance of the two downstream classification tasks. How do other data augmentation types impact the performance and results?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "This is a very well-written paper with clear organization of the figures and sound presentation of the data sets used, methods applied, and results obtained. The authors apply a successful contrastive learning method to develop a new framework that outperforms comparable state-of-the-art methods in both the cell-type classification task and the less widely-explored brain-region classification task. The task relies on two electrophysiological recording modalities: spiking activity and EAPs, which are more accessible, and the decent classification performances come with minimal fine-tuning. NEMO is able to differentiate between VIP and SST cells, which is highly valued in systems neuroscience, and its ability to yield data embeddings that lead to separable classification regions is impressive. The method described in this paper will be particularly helpful and useful to systems neuroscientists interested in applying this technique with the goal of decoding the neural circuitry underlying multiple biological functions." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents a novel application of contrastive learning to solve two important problems in systems neuroscience by incorporating just two electrophysiological recording modalities: spiking activity and extracellular action potentials (EAPs). The authors developed a new framework called Neuronal Embeddings via Multimodal Contrastive Learning (NEMO) by combining the well-established Contrastive Language-Image Pretraining (CLIP) framework with task-specific data augmentations and encoders. The authors demonstrate the multimodality as well as the power and utility of NEMO by evaluating its performance on two very different downstream tasks:\n\n1.\tcell-type classification among parvalbumin (PV), somatostatin (SST), and vasoactive intestinal peptide (VIP) inhibitory interneurons using opto-tagged Neuropixels Ultra (NP Ultra) recordings from the mouse visual cortex, and\n\n2.\tbrain-region classification among 10 broad areas using the public International Brain Laboratory (IBL) brain-wide map data set.\n\nThis paper’s novelty mainly stems from the utilization of a paired data set that combines an autocorrelogram (ACG) image of every neuron’s spiking activity and a waveform template of the neurons’ EAPs, and from the application of two separate encoders for the aforementioned two modalities. In both cell-type and brain-region classification tasks, the authors show that NEMO outperforms the state-of-the-art multimodal cell-type embedding methods including PhysMAP and VAE-based methods as well as a fully supervised method in terms of balanced accuracies and macro-averaged F1 scores with minimal fine-tuning." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Detailed graphical representations of the architectures used in the authors’ method may help the readers understand the details of NEMO better. A more comprehensive description, such as including explanations of 10D and 500D in the VAE baseline versions’ latent spaces of the encoder architectures, would have further aided clarity to the method’s explanations.\n\nThe authors restrict the number of example neurons, whose recorded activities are inputted to the overall architecture, to five neurons without an explanation of why the input neuron number was kept low. Providing a rationale for limiting the number of input neurons to five as well as an explanation of how the results of the authors' method change with varying input neuron number would be greatly appreciated.\n\nThe authors state that they fixed the hyperparameters for all methods across the two data sets used in the experiments due to the limited number of labels for evaluation for the cell-type classification task. It is unclear whether this choice led to fair performance comparisons among the state-of-the-art methods. It would help to know that separate hyperparameter optimization among the different methods and data sets would not yield different results. Providing a brief analysis on the sensitivity of the results and the overall performance to variations in specific hyperparameters will notably strengthen the claims made in this paper.\n\nMinor comments:\n\nThere seems to be a citation error or a missing preposition in Section 1 when the authors cite IBL et al.\n\nRadford et al. 2021 to Radford et al., 2021 in Section 4\n\nTable 6 is non-existent in Section 6.1. The authors may be referring to Table 1.\n\nThere seems to be a citation error or a missing preposition in Section 6.3 when the authors cite Chen et al. (2020).\n\nFigure 5 (b) caption's word \"then\" should be changed to \"than.\"\n\nIn Section 7, the phrase \"should also be useful these down-stream tasks\" should be changed to \"should also be useful in these down-stream tasks.\"" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1.The authors emphasized “in vivo”, however, are there any results about the computational efficiency of the NEMO model that can support it? Does it support real-time processing?\n\n2.I would expect a validation of the data augmentation strategy. It is understandable that the construction of ACG images is computationally expensive. But the authors are encouraged to validate the data augmentation strategy adopted in the study, augmentations directly for the ACG images, is reasonable by showing a couple examples.\n\n3.The authors compared NEMO with the VAE-based method. However, in my opinion, it seems that a critical comparison is missing: the comparison between naive models, specifically, NEMO without fine-tuning and the VAE-based method without fine-tuning. This comparison would highlight the representational power of the two methods ." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "As mentioned above, this study is among the earliest efforts to apply contrastive learning for joint modeling of extracellular action potential data and spiking activity data. Most key components in the proposed analytical framework are fetched from previous work (e.g., CLIP contrastive learning and the ACG encoder), increasing the reproducibility of the study. In addition, the study was well presented and easy to follow." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors proposed a multimodal contrastive learning approach for joint modeling of extracellular action potential data and spiking activity data. The proposed approach can be fine-tuned for different downstream tasks, including in vivo cell-type and brain region classification. More specifically, the author applied the classic contrastive learning framework established in CLIP on extracellular action potential data and spiking activity data. Although the theoretical innovation is relatively limited, the authors made the earliest efforts (as far as I know) to apply contrastive learning for joint modeling of extracellular action potential data and spiking activity data." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "As mentioned above, the theoretically contribution of the study is relatively limited. The readers may expect to see some components that are specifically designed with consideration for the unique characteristics of the data and the problem being addressed." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "- The VAE training is unclear to me. Do you jointly embed waveforms and autocorrelograms, or do you use two separate encoders/decoders with a shared latent space (which would require cross-decoding, i.e. encode waveform and decoder autocorrelogram)?\n- How important are the data augmentations? Can you provide an ablation experiment for that?\n- Waveforms and autocorrelograms are two reasonable choices for input modalities. However they are not the only conceivable choices. Have you thought/tried other choices or thought about learning an encoding of spiking activity directly?\n- What are the results when you cluster on the latent embeddings directly instead of running UMAP first? How stable is the clustering? I.e. if you train two models of NEMO from different seeds and then cluster neurons, how consistently are two neurons assigned to the same cluster (as measured by adjusted rand index or similar)?\n- Can you provide a definition for the modularity index?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "- The paper is well written, the figures are well made, descriptions are generally clear.\n- The paper contains extensive additional material for more details.\n- The training and experimental setup seems to be carefully chosen and sound.\n- Limitations are discussed at the end." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces NEMO (Neuronal Embeddings via MultimOdal contrastive learning), a method for classifying neurons by their cell type and brain region location using electrophysiological data. NEMO uses CLIP-like contrastive learning to jointly analyze two types of neural data: the shape of neural electrical signals (waveforms) and patterns of neural activity over time (autocorrelograms). \n\nThe authors evaluated NEMO on two datasets: an opto-tagged mouse visual cortex dataset for cell-type classification and the International Brain Laboratory dataset for brain region classification. In comparative experiments, NEMO achieved higher classification accuracy than existing methods including PhysMAP and VAE-based approaches. The authors also demonstrated that using multiple units improved performance, and that the method maintained effectiveness even with limited labeled training data. The paper includes ablation studies examining the importance of joint versus independent learning of the two data modalities." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "I only found minor weaknesses. \n\n- Although, the authors did a great job providing necessary details, sometimes the training specifics for the control models and the clustering analyses were a bit too short in my opinion. It would be great if you could provide a bit more detail on this.\n- It would be great to see an ablation for the data augmentation to see how important that is (see questions).\n\n**Minor (do not expect you to respond to this)**\n\n- Figure 3 typo in legend “Supervise” instead of “Supervised”\n- Supplementary Figure 7 is blurred likely due to plotting settings\n- As multi-unit has a particular meaning in neuronscience, I find the wording “multi-unit brain region classification”. I think you mean “multi-neuron” here. Unless I misunderstood and you are really using multi-unit activity, I would change the wording." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. In Figure 1a, the text \"Neuropixels 1.0\" should be replaced with \"Neuropixels Ultra\" since your VIP/SST/PV data comes from NP Ultra, not NP 1.0, which is used in the IBL dataset for classifying brain regions. Also, please update the inset to show the schematic of NP Ultra instead of NP 1.0.\n\n2. Figure 3b is not mentioned in the text. It could be referenced between L350 and L361.\n\n3. Figure 3c is also not mentioned in the text. It might fit well between L362 and L366.\n\n4. Consider changing the title of Figure 3b and 3c from \"unit\" to \"neuron\" to be consistent with the terminology used in the main text.\n\n5. In the PhysMAP paper that you benchmarked, they applied three public datasets from S1, A1, and V1/Hippocampus. Have you tested these datasets? While I am not requesting additional experiments, if you have tested them, it would be helpful to include the results in the supplementary materials.\n\n6. Please specify how you computed the two primary metrics: macro-averaged F1 score and balanced accuracy. Does the first metric equal the unweighted mean of all the per-cell-type F1 scores? Does the second metric equal the unweighted mean of sensitivity (True Positive) and specificity (True Negative)?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The writing is clear, easy to understand, and follows a smooth logical flow, with almost no typos and well-presented figures.\n\n2. The paper provides a thorough review of relevant literature in this field. I have closely followed the cell-type classification area, and all the papers I am aware of (and some I was not) have been accurately cited, except for one (see Weakness). Notably, the authors benchmarked two very recent models that are still in preprint format.\n\n3. This is the second work to use contrastive learning for cell-type classification and the first to combine two modalities.\n\n4. The multimodal approach outperforms single-modal models, which is consistent with previous studies." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors used multimodal (spike waveform + firing pattern) contrastive learning to classify three types of inhibitory neurons (PV/SST/VIP) and ten different brain regions. The proposed NEMO model is based on the widely used CLIP (Contrastive Language-Image Pre-Training) model. NEMO outperforms two previous multimodal models: PhysMAP and VAE." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. In the best-case scenario (cell type classification, L344, Figure 2c), NEMO outperforms VAE by 11%. However, in brain region classification, the improvement is minimal. For example, in Figure 3e, comparing the deep orange (NEMO) to the deep blue (VAE) bars, the difference in balanced accuracy is less than 0.05. Are these differences statistically significant?\n\n2. Joint training shows little to no benefit over independent training. For example, in Figure 5b, comparing the deep orange (NEMO) to the deep violet (SimCLR) bars, the difference in balanced accuracy is around 0.01. Additionally, in Supplementary Table 9, independent NEMO performs either better (0.84 vs. 0.83) or similarly (0.83 vs. 0.84, 0.87 vs. 0.88) to joint NEMO. Are these differences statistically significant?\n\n3. To my knowledge, there is no neuroscience evidence suggesting a strong pairwise correlation between spike waveform and firing patterns. For example, layer 5 pyramidal neurons and cortical PV neurons both fire a large number of spikes (both spontaneously and evoked), but their spike waveforms are broad and narrow, respectively (Cortical connectivity and sensory coding, KD Harris, TD Mrsic-Flogel - Nature, 2013). Additionally, burst firing can be evoked in both excitatory (Chattering cells: superficial pyramidal neurons contributing to the generation of synchronous oscillations in the visual cortex. CM Gray, DA McCormick - Science, 1996) and SST neurons (Somatostatin-expressing neurons in cortical networks, Urban-Ciecko, AL Barth - Nature Reviews Neuroscience, 2016). This is fundamentally different from the relationship between an image and its description in CLIP. In other words, the word \"puppy\" and an image of a \"puppy\" represent the same concept, but a broad spike could be associated with either burst or dense firing, depending on whether the neuron is located in layer 2/3 or layer 5.\n\n4. This is the second paper to use contrastive learning for cell-type classification. The first is CEED (Vishnubhotla et al., 2023), which used an unsupervised model (SimCLR) to classify cell types and benchmarked against WaveMap (the single-modality version of PhysMAP). While the CEED work does not significantly compromise the novelty of this study, it should be more clearly acknowledged. The current citation, \"Contrastive learning has been applied to raw electrophysiological recordings (Vishnubhotla et al., 2024),\" is inappropriate." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024in,\ntitle={In vivo cell-type and brain region classification via multimodal contrastive learning},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=10JOlFIPjt},\nnote={under review}\n}" }, "abstract": { "value": "Current electrophysiological approaches can track the activity of many neurons, yet it is usually unknown which cell-types or brain areas are being recorded without further molecular or histological analysis. Developing accurate and scalable algorithms for identifying the cell-type and brain region of recorded neurons is thus crucial for improving our understanding of neural computation. In this work, we develop a multimodal contrastive learning approach for neural data that can be fine-tuned for different downstream tasks, including inference of cell-type and brain location. We utilize multimodal contrastive learning to jointly embed the activity autocorrelations and extracellular waveforms of individual neurons. We demonstrate that our embedding approach, Neuronal Embeddings via MultimOdal Contrastive Learning (NEMO), paired with supervised fine-tuning, achieves state-of-the-art cell-type classification for an opto-tagged visual cortex dataset and for brain region classification of the public International Brain Laboratory brain-wide map dataset. Our method represents a promising step towards accurate cell-type and brain region classification from electrophysiological recordings." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "contrastive learning", "electrophysiology", "extracellular", "multimodal", "neuroscience", "cell type", "brain region", "Neuropixels", "deep learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/430bdaf44a0512940ae8575e2683f9503e0bc3a0.pdf" }, "presentation": null, "primary_area": { "value": "applications to neuroscience & cognitive science" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "In vivo cell-type and brain region classification via multimodal contrastive learning" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "- The idea of studying the effects of prompts through the geometry of latent representations is interesting." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper investigates the effect of prompts on the representation of the EOS token across prompts and LLMs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Studying only the representation of the EOS token seems to be an oversimplification. While this representation technically can depend on all of the context/generation, it might not attend to its meaningful parts, thus failing to capture interesting patterns.\n- The paper lacks clear/practical insights besides observing that the EOS token representation does depend on the prompt in some way.\n- For an empirically oriented work, studying only binary sentiment classification datasets is insufficient. Hypotheses should be verified across a broader range of tasks, including open-ended generation.\n\nOther:\n- Many typos throughout the paper (missing spaces and periods, inconsistent usage of citet and citep, etc.)\n- IsoScore is not defined in the paper. Only a high-level explanation is provided.\n- RIS is not carefully defined - is k' the same as c? \"value of k is equal to itself\" is not a clear statement." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "(1) What are the motivations of grouping the prompts and how the groups of prompts correlate to the focus of the model generation? Why do the authors use the last layer rather than the embedding layer for clustering the prompts? \n\n(2) What is the mathematical formulation of the Isoscore? What does it used for in the experiments?\n\n(3) Several studies, such as PromptRobust [1], focus on evaluating prompt robustness. How does this approach offer advantages over existing methods?\n\n(4) The paper offers limited technical contributions and focuses narrowly on the classification task. Could the work be applied to generation task? \n\n[1] Zhu, K., Wang, J., Zhou, J., Wang, Z., Chen, H., Wang, Y., ... & Xie, X. (2023). Promptbench: Towards evaluating the robustness of large language models on adversarial prompts. arXiv preprint arXiv:2306.04528." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "The paper tries to address the intriguing question of how prompt representations relate to model performance, with a specific focus on how prompt formulation impacts performance robustness." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper investigates how prompts affect the intrinsic geometry of LLMs. The authors explore two research questions: (1) whether prompts alter the intrinsic dimensionality of representations, and (2) whether prompts can be grouped by their impact on model performance and vector representations using clustering methods." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The paper's format is somewhat disorganized, with unexpected line breaks and misplaced punctuation. Also, some figures like figure 3 do not have illustrations, making it hard to understand." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "- Why are we interested in isotropy of the EOS token hidden states?\n\n- Can we use the isoscore as a selection criterion for picking a good prompt? This possibility is hinted at in the introduction \"Our objective is to establish a direct correlation between the prompts, latent representations, and the model performance.\" but not deeply explored. The results sound negative on this front---\"we cannot link the IsoScore to the performance of the model and prompt\". How does this finding relate to the cited works that show isotropy is a good property for embedding models?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper studies a general and interesting question of how LLM internals (such as hidden states) evolve/change when certain parts of the input (such as the prompt) change. Further understanding of this process would be useful for several downstream applications, such as (as shown in prior work) detecting adversarial prompt injection attacks." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies how variations in the prompt change the distribution of hidden states in LLMs for zero-shot binary classification. The last hidden state representation is extracted at each layer for a variety of different prompts. The authors plot how the IsoScore of the hidden states evolves across layers. These hidden states are then clustered, based on which the authors conclude that \"[the] clustering tends to focus on other characteristics than the prompt itself.\"" }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The biggest issue with the draft is the presentation of the results. There are numerous spelling, writing, and formatting errors, a subset of which I listed below. The first two paragraphs of the introduction are unnecessary background for the ICLR audience. The related work section on LLMs is not particularly related to the focus of this paper (understanding how prompts change internal model representations). That section is missing citations to LogitLens and its derivatives (e.g. https://arxiv.org/pdf/2303.08112), which also study how changes in the prompt change model internals. For example, that paper develops classifiers (based on features extracted from model internals) for detecting prompt injection. I.e., they already studied this paper's \"hypothesis 2\" that \"The geometrical characteristics are sufficiently discriminating to facilitate the grouping or separation of prompts and comprehend how the model processes them (HP2).\" The details of very relevant parts of the paper, such as a brief description of the IsoScore algorithm, are missing. Isotropy is never formally defined and the draft never explains why it is a desirable property for decoder-only models (beyond references to other work studying isotropy in the context of embedding models). The details of precisely how the IsoScore is computed are missing to the point where I'm not sure the results of the paper are reproducible.\n\nI think minor writing and presentation issues are not disqualifying, but in this case they are pervasive and make it difficult to judge the technical aspects of the paper on merit.\n\nSubset of writing issues:\n- L16 \"heuristics rules\" --> \"heuristic rules\"\n- L24 \"their impact\" --> \"its impact\"\n\n- first two paragraph of the intro are unnecessary bg for the ICLR audience.\n\n- L53 starts \"However\" but is not contradicting a previous point\n\n- L67 what does \"intrinsic\" mean here?\n- L67 double period\n- L69 missing period.\n\n- L74 \"LLMs representations leveraging prompting\" --> \"LLM representations that leverage prompting\"\n\n- L101 \"section 4\" --> \"Section 4\".\n\n- L123, 137, L286 have \\citet that should be \\citep\n\n- L174 \"information, Therefore\" --> \"information. Therefore,\"\n\n- L178 \"Dimensionnality\"\n\n- L184 \"PCA get some limitations\"\n\n- L214-L215 $k$ not in latex formatting\n\n- L284 \"We describe, as follow,\"\n\n- L290, L306 \"for instance :\", \"be written :\"\n\n- L303 \"we constraint the model\"\n\n- L317 \"First analyse results of isoscore are discussed\"\n\n- L322 \"Since IsoScore is a recent algorithms\"\n\n- L323 \"accross\"\n\n- L483 \"This allows us give answers\"" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "Geometrical analysis of prompt impact on latent representation in LLMs" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024impact,\ntitle={Impact of Prompt on Latent Representations in {LLM}s},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=10kBEqYKKN},\nnote={under review}\n}" }, "abstract": { "value": "The effectiveness of zero-shot learning frameworks, particularly in Large Language Models (LLMs), has lately shown tremendous improvement. Nonetheless, zero-shot performance critically depends on the prompt quality. Scientific literature has been prolific in proposing methods to select, create, and evaluate prompts from a language or performance perspective, changing their phrasing or creating them following heuristics rules. While these approaches are intuitive, they are insufficient in unveiling the internal mechanisms of Large Language Models. In this work, we propose exploring the impact of prompts on the latent representations of auto-regressive transformer models considering a zero-shot setting. We focus on the geometrical properties of prompts' inner representation at different stages of the model. Experiments conducted give insights into how prompt characteristics influence the structure and distribution of vector representations in generative models. We focus on binary classification tasks on which prompting methods have shown robust performance and show that prompt formulation has indeed an influence on latent representation. However, their impact is dependent on the model family. Using clustering methods, we show that even though prompts are similar in natural language, surprisingly, their representations can differ. This is highly model-dependent, demonstrating the need for more precise analysis." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Explainability", "Representation analysis", "LLM", "prompting", "zero-shot" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/632c9289cdb828739212f20f08e123d679b624e5.pdf" }, "presentation": null, "primary_area": { "value": "interpretability and explainable AI" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Impact of Prompt on Latent Representations in LLMs" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "1. The paper focuses on and attempts to address a crucial yet highly challenging problem in the field of graph analysis—constructing a Graph Foundation Model (GFM).\n2. On commonly used graph datasets, the model OMOG presented in the paper achieves relatively good performance in both zero-shot and few-shot settings." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes the OMOG (One Model for One Graph) framework, which advances graph learning by pre-training a unique model for each graph within a model bank. By creating a bank of expert models, each pre-trained for a specific dataset, OMOG selects relevant experts for inference using gate modules tailored to the target graph’s domain. This approach mitigates negative transfer issues common in multi-domain pre-training, and it performs effectively in zero-shot and few-shot tasks like link prediction and node classification, showing its potential for cross-domain graph tasks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper does not clearly explain the differences from other MOE-based methods, such as GraphAlign and AnyGraph. The approach seems very similar to these methods, leaving it unclear what specific advantages OMOG has over them and why it achieves improved performance.\n2. A core idea of OMOG is that each pre-training dataset requires a dedicated expert. This approach poses challenges for scalability: as the volume of pre-training data increases, the model grows linearly with the data, which is detrimental to pre-training efficiency.\n3. Why is the expert model specifically a Transformer? How would the performance change if other models, such as GNN, Graph Transformer, or MLP, were used instead? Additionally, prior to entering the experts, the features and structure are fused through SGC. Why couldn’t this fusion step be incorporated within the experts themselves? After all, different graphs may require varying levels of neighbor aggregation.\n4. The core part for achieving zero-shot in this paper relies on calculating the similarity between label embeddings and prediction embeddings to obtain the final label prediction. In fact, most models that work under few-shot settings can be adapted to zero-shot using a similar approach. Consequently, Table 1 lacks several relevant baselines, such as GraphAlign, GCOPE, and GraphMAE.\n5. Do all experts contribute to downstream performance improvements? In Figure 6, while the number of experts is adjusted, the full set of pre-training data is still used to train the gating mechanism. Could you vary the number of pre-training datasets to examine how this affects downstream performance?\n6. Although this paper discusses GFM, which should be applicable to various downstream tasks, there is still an absence of experiments on graph-level tasks, such as graph classification or graph regression.\n7. Some parts of the paper lack clarity. For example, in Section 4.5, the phrase ‘select 10 samples from each dataset’ is ambiguous. Does this refer to selecting 10 nodes, subgraphs, or something else?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. What is the time complexity or running time of the proposed method given that the proposed method needs to pretrained the model on several graphs from different domains?\n2. In line 207, the authors mention that the node-level feature are randomly maked, resulting in two maked views. Are these two masked views mutually exclusive? For instance, given a 10-d feature matrix, the first masked view is generated by masked 5 features and the second view is generated by masking the rest 5 features?\n3. How do you ensure that the generator only generates a matrix to mask the domain irrelevant features such that the filter features are domain-related?\n4. The authors mention that one key issue is negative transfer. Since the knowledge in the proposed methods is extracted from graphs in different domains, it inevitably increases the probability of facing the negative transfer issue. How does the proposed method address this issue? The top-k strategy seems to only filter out the low confidant knowledge, while it can not directly alleviate the negative transfer issue as the irrelevant knowledge might be included in the top k expert models. \n5. I try to reproduce the experimental results, but there is no instruction and datasets available in the provide GitHub link. The readme seems to be empty. Could you provide the datasets and the instruction to reproduce the results?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The presentation of this paper is good and most parts of the paper are clear.\n2. This paper proposes a novel cross-domain pretraining framework.\n3. The experimental results demonstrate the effectiveness of the proposed method." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a novel cross-domain pretraining framework called \"one model for one graph,\" by pretraining a bank of expert models and using a gating function to choose a subset of experts to effectively integrate prior model knowledge." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The intuition of the generator and filter in Pretraining the gate module is not clear.\n2. The authors lack the discussion about the difference between the proposed method and the mixture-of-experts based methods. \n3. I am concerned about the negative transfer issue in the proposed method. Since the knowledge in the proposed methods is extracted from graphs in different domains (and most of them are irrelevant), it inevitably increases the probability of facing the negative transfer issue. How does the proposed method address this issue? The top-k strategy seems to only filter out the low confidant knowledge, while it can not directly alleviate the negative transfer issue as the irrelevant knowledge might be included in the top k expert models. \n4. I try to reproduce the experimental results, but there is no instruction and datasets available in the provide GitHub link." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "see Weaknesses" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1.\tBy pretraining individual models for each graph dataset, OMOG effectively addresses the feature and structural heterogeneity found across diverse graphs.\n2.\tOMOG’s model bank allows the easy addition of new expert models without retraining the entire system, providing flexibility to expand the pretraining bank with new data and adapt quickly to novel domains." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes OMOG, an innovative cross-domain graph learning framework designed to enhance the adaptability and performance of Graph Neural Networks (GNNs) across various domains. By training a distinct expert model for each pre-training graph and employing adaptive gating functions during inference, OMOG dynamically selects relevant experts for unseen graphs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tThe primary motivation for adopting the “one model for one graph” approach is to alleviate the negative transfer limitations observed in the “one model for all graphs” method. It would be beneficial to provide comparisons and discussions on how this method differs from prior approaches that aim to reduce negative transfer through better pretraining data selection as [1].\n2.\tIt is recommended to identify which models, pretrained on specific graphs, are selected as the best match for various test graphs, with explanations for these selections. Additionally, I’m curious about whether pretraining data from different domains can contribute effectively or if only similar/same-domain data is more beneficial would strengthen the analysis. A case study is recommended to evaluate whether the proposed gating strategy actually mitigate issues stemming from conflicts across pre-training data from diverse domains.\n3.\tIt would be valuable to explore whether this pipeline could be adapted to other self-supervised learning approaches for pretraining graph-specific experts, and additional ablation studies are expected.\n4.\tOMOG’s design requires a separate model for each dataset and can result in a large model bank when many datasets are involved. Will it lead to high storage costs and maintenance overhead, especially in resource-constrained environments? A complexity analysis would also be helpful to understand OMOG’s computational feasibility at scale.\n\n[1] Better with less: a data-active perspective on pre-training graph neural networks. In NIPS '23." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. What is the means of t in Equation (3)? How do you ensure that each expert has different parameters? Based on Equation (3), it seems that each expert has the same input and parameters, leading to identical outputs. What, then, is the purpose of having multiple experts?\n2. In line 241, the mask matrix seems to be replaced by an MLP. Why are the node embeddings transformed by the MLP considered to be negative embeddings?\n3. In Equation (4), is f_center ​the average output of a single graph across multiple experts, or the average output of multiple source graphs through their respective experts? How does this function as an anchor point for training the gate?\n4. What are the parameters of the Gate in Equation (5)? Why is the Gate used before the Expert?\n5. According to the inference process, an unseen graph activates the top k experts based on the highest correlation between each expert's output and the output average. How do you ensure that the pre-trained graph domain encompasses a sufficiently heterogeneous feature space to handle all potentially unseen graph domains?\n6. How is the total number of experts determined, especially when the number of graph domains during pre-training and testing is uncertain?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The approach of building an expert model for each graph intuitively addresses the issue of negative transfer in graph pre-training.\n2. The problem addressed in this paper is a crucial part of foundational research on graph models and is currently a topic of significant interest among researchers.\n3. The experiments involve graph data from multiple domains and include comparisons with recent, noteworthy methods in graph transfer learning." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes OMOG to pretrain one model for one graph in cross-domain transformation. OMOG uses SGC as message aggregation before experts learning, and uses the contrastive method for expert and gate training. In the inference stage, OMOG activates the top-k of experts to infer node labels. The experimental results proved the effectiveness of the method." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. According to the method statement, graphs from different domains are required to have input spaces of the same size, which seems difficult to satisfy with real-world data.\n2. The construction of multiple experts for input graphs appears to be relatively naive, as it merely involves repeating several encoders and using similarity ranking with a central vector for averaging activation." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "The work proposes a new framework to boost knowledge transfer during cross-domain graph pretraining." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024one,\ntitle={One Model for One Graph: A New Perspective for Pretraining with Cross-domain Graphs},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=10vaHIOdEe},\nnote={under review}\n}" }, "abstract": { "value": "Graph Neural Networks (GNNs) have emerged as a powerful tool to capture intricate network patterns, achieving successes across different domains. However, existing GNNs require careful domain-specific architecture designs and training from scratch on each dataset, leading to an expertise-intensive process with difficulty in generalizing across graphs from different domains. Therefore, it can be hard for practitioners to infer which GNN model can generalize well to graphs from their domains. To address this challenge, we propose a novel cross-domain pretraining framework, \"one model for one graph,\" which overcomes the limitations of previous approaches that failed to use a single GNN to capture diverse graph patterns across domains with significant gaps. Specifically, we pretrain a bank of expert models, with each one corresponding to a specific dataset. When inferring to a new graph, gating functions choose a subset of experts to effectively integrate prior model knowledge while avoiding negative transfer. Extensive experiments consistently demonstrate the superiority of our proposed method on both link prediction and node classification tasks." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Graph Pretraining; Cross-domain Graph Learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/d68cf29a660bd027219ea8dc1f02e556d61af29d.pdf" }, "presentation": null, "primary_area": { "value": "learning on graphs and other geometries & topologies" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "One Model for One Graph: A New Perspective for Pretraining with Cross-domain Graphs" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "I wonder for items like causality and new synthesis if an approach more configurable could take place now that they have separated the pedestrians from the road. \n\nThinking of something like this \nWang, Cheng Yao, et al. \"CityLifeSim: A High-Fidelity Pedestrian and Vehicle Simulation with Complex Behaviors.\" 2022 IEEE 2nd International Conference on Intelligent Reality (ICIR). IEEE, 2022.\n\nWhere the data is later attached to an engine like Carla or airsim" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "Comprehensive Dynamic Modeling: OmniRe can handle various actors in urban settings, unlike most previous methods that focus mainly on vehicles.\n\nScene Graphs and Gaussian Splatting: The system uses 3D Gaussian splatting for detailed scene and object rendering, including control over each object.\n\nHuman Behavior Simulation: Through SMPL modeling, OmniRe accurately reconstructs human motions, even in cluttered environments, enabling simulations of interactions between pedestrians and vehicles.\n\nState-of-the-Art Performance: Extensive testing on datasets like Waymo and others show OmniRe significantly outperforms existing methods in terms of visual fidelity and reconstruction accuracy." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "OmniRe is a framework designed to create high-fidelity digital twins of dynamic urban scenes for simulations, particularly for applications in autonomous driving. OmniRe goes beyond vehicle modeling to support diverse dynamic actors like pedestrians and cyclists, enabling complex simulations that reflect real-world scenarios. It utilizes Gaussian Scene Graphs with multiple representations, allowing detailed and editable scene reconstructions with both rigid (e.g., vehicles) and non-rigid (e.g., pedestrians) actors." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "There are several limitations in OmniRe approach, which are correctly identified in the paper too.\n\nLighting Effects: OmniRe doesn’t model lighting variations explicitly. This can lead to visual inconsistencies when combining scene elements with differing lighting conditions, which may reduce realism in certain simulations. Addressing this would require additional modeling of lighting dynamics.\n\nNovel View Synthesis Limitations: OmniRe’s per-scene optimization approach struggles to generate satisfactory results when the camera view deviates significantly from the training trajectories. This could be a limitation for scenarios requiring a wide range of viewing angles, such as free navigation through the reconstructed scenes. The authors suggest incorporating data-driven priors or generative models as future work to address this.\n\nComputational Complexity: While the method achieves high-quality reconstructions, the complexity of the Gaussian Scene Graph and the joint optimization of multiple parameters (pose, appearance, etc.) require substantial computational resources. Training time per scene, though optimized for an RTX 4090 GPU, could still pose scalability issues for large datasets or continuous real-time simulation needs.\n\nChallenges with Real-Time Adaptability: The method’s reliance on SMPL modeling for human actors and per-node deformation fields, though effective, might introduce delays in real-time applications, particularly if scenes are highly dynamic or involve many non-rigid actors." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "Please refer to the Weaknesses." }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "1. The paper is well-organized and easy to follow.\n2. The proposed method for in-the-wild human representation is straightforward yet crucial for driving scene reconstruction.\n3. Both quantitative and qualitative experiments effectively support the claims made in the introduction, with OmniRe achieving state-of-the-art results across various experimental settings." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces OmniRe, a novel approach for urban scene reconstruction that focuses on dynamic actors, including vehicles, pedestrians, and cyclists. OmniRe employs a Gaussian Scene Graph-based framework to model both static and dynamic objects. To address the limitations of previous methods in reconstructing non-rigid human models, OmniRe integrates SMPL for in-the-wild human representation, allowing for joint-level control. Extensive evaluations across several driving datasets demonstrate OmniRe's superior performance compared to baseline methods." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Handling Occlusions and Complex Dynamics: OmniRe addresses in-the-wild challenges, yet the performance might be limited by severe occlusions and overlapping actors in complex urban scenes. Further refinement or integration of advanced occlusion handling techniques could enhance reconstruction fidelity.\n2. Performance in Specific Urban Scenes: For specialized scenarios, such as highways (with fast-moving vehicles), nighttime environments, and adverse weather conditions, does OmniRe maintain high reconstruction quality under these challenging conditions?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. GS-based methods generally perform well in scenarios with static environments or low vehicle speeds, as demonstrated by most of the demos on the project page. However, I am curious about the reconstruction performance of this approach in situations where the ego vehicle is moving at higher speeds.\n2. I wonder about the computational cost of reconstructing a complete segment in the Waymo dataset, as the entire pipeline seems a bit complex.\n3. Why does it seem that the reconstruction quality of NuPlan is significantly worse than that of other datasets?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper is clearly written, with illustrative figures that are easy to understand. The experiments are comprehensive.\n2. Modeling dynamic objects and simulating interactive behaviors are essential for closed-loop simulation in autonomous driving systems.\n3. This work is highly engineering-oriented and demonstrates impressive results. Additionally, the authors have committed to open-sourcing the code, which will have significant value in advancing autonomous driving simulation in the future." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces OmniRe, a comprehensive framework for dynamic urban scene reconstruction. It leverages neural scene graphs with Gaussian representations to unify the reconstruction of static backgrounds, moving vehicles, and non-rigidly dynamic actors. Additionally, it incorporates specialized designs for human modeling. The effectiveness of the approach is demonstrated across multiple datasets, showcasing superior performance in both reconstruction quality and novel view synthesis." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "As mentioned by the authors in the limitations section, there are still two key shortcomings: 1. The lack of lighting modeling results in unnatural object insertions. 2. The synthesis of new viewpoints is constrained to the original trajectory, limiting the approach from achieving fully free-trajectory digital reconstruction." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024omnire,\ntitle={OmniRe: Omni Urban Scene Reconstruction},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=11xgiMEI5o},\nnote={under review}\n}" }, "abstract": { "value": "We introduce OmniRe, a comprehensive system for efficiently creating high-fidelity digital twins of dynamic real-world scenes from on-device logs. Recent methods using neural fields or Gaussian Splatting primarily focus on vehicles, hindering a holistic framework for all dynamic foregrounds demanded by downstream applications, e.g., the simulation of human behavior. OmniRe extends beyond vehicle modeling to enable accurate, full-length reconstruction of diverse dynamic objects in urban scenes. Our approach builds scene graphs on 3DGS and constructs multiple Gaussian representations in canonical spaces that model various dynamic actors, including vehicles, pedestrians, cyclists, and others. OmniRe allows holistically reconstructing any dynamic object in the scene, enabling advanced simulations (~60 Hz) that include human-participated scenarios, such as pedestrian behavior simulation and human-vehicle interaction. This comprehensive simulation capability is unmatched by existing methods. Extensive evaluations on the Waymo dataset show that our approach outperforms prior state-of-the-art methods quantitatively and qualitatively by a large margin. We further extend our results to 5 additional popular driving datasets to demonstrate its generalizability on common urban scenes. We will make the code and data publicly available." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Gaussians Splatting", "Neural Rendering", "Dynamic Scene Reconstruction", "Autonomous Driving" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/1b615681674ee6446128cbfd3f1fe6afc4ce2693.pdf" }, "presentation": null, "primary_area": { "value": "applications to robotics, autonomy, planning" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/a6075c5ab9b491cfda578768d4b601e58dc6dc5b.zip" }, "title": { "value": "OmniRe: Omni Urban Scene Reconstruction" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "My primary suggestion for strengthening this paper is more or less solely for the authors to lean further into its greatest strength -- and to further explicate or justify the expert methods that differentiate this work from so many others attempting to tackle similar questions. Needless to say, perhaps, there are a number of ways the authors could do this. Below are a few different “options” that (I hope) seem reasonable given the constraints of the current review. The authors should feel free to choose however many / whichever of these seems most feasible or intuitive. For me, at least, almost any movement along these vectors is movement that would increase the value of this work for the target audience it seems intended for:\n- “Beyond accuracy”: The primary justification of the authors’ “novel readout mechanism” is the general increase in accuracy it provides over other methods. But the emphasis on accuracy as the primary advantage rings a bit hollow if a major part of the goal here is to gain insight into the structure of representation in biological cortex. There are many alternative ways (e.g. data augmentation, denoising, nonlinearities) -- even “hacky” ones (e.g. smaller cross-validation splits) that one could use to increase the predictive accuracy of model readout mechanisms. What demonstrable advantage does the “semantic spatial transformer” readout have over readout methods with respect to the theoretical questions at play here? (An example answer: “ordinal least squares or regularized regression-style readouts do not preserve spatial information -- therefore making certain areas of the brain appear to be more transform-invariant than they likely are in reality. Here’s a metric that operationalizes the probability of transform-invariance in the fMRI data without models. And here is a side-by-side comparison of the transform-invariance we estimate with ridge regression and our STN readout, respectively.”)\n- “Semantics” without language model confounds: There are a number of issues (again, beyond the scope of this paper, but nonetheless relevant) with the use of language models as predictive models of visual fMRI data -- including the fact the inputs to these models (tokenized words) are already proto-symbolic at the time of their initial injection into the candidate representational models that embed them (and are thus more abstract by default than the pixels injected into vision models); and also, an increasing “convergence” between vision and language models [1] that suggests a sort of “default” alignment between these systems attributable (most likely, it seems) to biases in their training data. How to get at questions of “semantics” without over-interpreting language models? One way, perhaps, is to reconsider the brain data itself: It has been suggested by [2] that certain kinds of transformations on the underlying brain data to be modeled (e.g. aggregating across multiple neurons in the case of electrophysiology) can instantiate properties like linearly-separable category boundaries not otherwise apparent without those transformations. If something like this is done on the features of vision models (e.g. averaging across multiple images of the same visual concept), do vision models begin to look more “semantic”? Perhaps the semantic spatial transformer could be used to unveil precisely the kinds of feature transformations that occur along the gradient from early to late visual cortex.\n- “densifying” the “single” captions: The authors claim that the localized semantic descriptions inherent to their “dense” captioning method unveil a noticeable midpoint between early, more spatiotopic representations and later, “globally” abstract representations. But is this really about the local tagging of an image’s subparts? Providing more comprehensive “single captions” of the full image that includes more extensive specification of details might close the gap between the dense captioning method and the global captioning -- but in a way that obviates the need to manually subdivide the image. In short, adding further detail (with and without explicitly spatial language) seems like an important control for downstream interpretation of this result.\n\n[1] Huh, M., Cheung, B., Wang, T., & Isola, P. (2024). The platonic representation hypothesis. arXiv preprint arXiv:2405.07987.\n[2] Vinken, K., Prince, J. S., Konkle, T., & Livingstone, M. S. (2023). The neural code for “face cells” is not face-specific. Science Advances, 9(35), eadg1736." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The use of deep neural network models to predict and understand the structure of representation in the biological visual system is a practice rife with heretofore unanswered, but deeply foundational questions as to how it should be done. Bucking a trend that far too often recycles canonical, but relatively unscrutinized methods to new models or new brain data, this submission is impressive not just for the fact that it tackles these questions head-on, but tackles so many of them simultaneously -- and does so (mostly) without losing the forest for the trees. For this alone, I applaud the authors and can recommend that this paper be accepted." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this work, the authors present a comprehensive suite of analyses comparing vision / language DNN models to human fMRI data. Using a novel “readout” mechanism designed explicitly to account for space in the mapping of DNN embeddings to brain activity, the authors report localizing 3 sub-regions in the human visual cortex that respond differentially to spatial and semantic information." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "My major concern here (and one that I admit is not fully within the authors control, but which clarifying updates or different narrative focus could nonetheless address) is the lingering doubt as to whether even these newer, more expertly designed methods actually do give us any meaningful new “insights” about the biological system they’re nominally designed to give us insights about. An overly reductionist summary of the “findings” of this analysis with respect to the human visual brain could well be that they simply provide more evidence for what is already a amply established gradient of increasingly “abstract” visual information from early (more view-dependent) areas (where smaller, localized receptive fields and retinotopy are the dominant representational motifs) to later (less view-dependent) visual areas (where -- depending on which side of the ventral / dorsal divide those areas are closer to -- you begin to get “representations” that evoke “object categories”, “navigational affordances”, or “conceptual semantics”). And while much debate does remain as to many of the details here, it seems (to me at least) that the existence of this gradient is more or less a common consensus." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "Please see weaknesses section for the questions." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "I think overall, the authors' thorough experimentation is the greatest strength of this paper:\n\n* **Systematic Comparison:** They do a reasonably systematic comparison, comparing a diverse range of models and readout mechanisms, which offers valuable insights.\n* **Novel Readout Mechanism:** They propose (in the context of fMRI encoders) a novel readout mechanism—the using the previously proposed spatial transformer with differentiable bilinear sampling —and show that it indeed improves prediction accuracy. This is a significant contribution in the context of fMRI encoders.\n* **Identification of Cortical Regions:** They identify three cortical regions, largely aligned with prior hypotheses about visual cortical functionality. This further strengthens existing theories.\n* **Good Discussion of Prior Work:** The authors do a reasonably good job in discussing prior fMRI encoder literature, effectively contextualizing their research. This demonstrates a solid understanding of the field." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "Using fMRI data from the Natural Scenes Dataset, the authors investigate how different encoder backbones and readout mechanisms predict neural responses in the human visual system. \n\nThey compare a range of models, including those optimized for visual recognition (e.g., AlexNet, ResNet), neural response prediction, and language or vision-language tasks (e.g., CLIP, MPNET). Furthermore, they explore various readout mechanisms to map model activations to fMRI signals, introducing a novel approach (in the context of fMRI encoders) called the Semantic Spatial Transformer readout.\n\nThey find that:\n\n1. Response-optimized models perform best in early visual areas (V1-V4): This suggests that these areas prioritize perceptual features not readily captured by linguistic descriptions or task-specific training.\n\n2. Task-optimized and language models do better in higher visual areas: This indicates a shift towards semantic processing in these regions. Large language model embeddings, particularly those using detailed contextual descriptions, prove highly effective.\n\n3. Semantic Spatial Transformer readout improves performance: This novel readout consistently outperforms existing methods like linear, Gaussian, and factorized readouts, boosting accuracy by 3-23%. This improvement stems from its ability to learn stimulus-specific modulations of receptive fields and feature maps." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The presentation of this paper could be *significantly* improved. I think the presentation quality of this paper does not match the quality of other ICLR papers I am currently reviewing or have reviewed in past years, or ICLR papers that have been accepted in prior years.\n\nThe figures are unclear and lack consistent formatting, notation often unexplained, and significant wasted space.\n\nMy specific concerns are below:\n1. Figure 1 -> This figure is very cluttered and very confusing. Why are the subfigure legends (A, and B) placed so randomly?\n2. Figure 1 -> What is the 'What' weight matrix, where does it come from? It is obvious if you are familiar with fMRI encoder literature, but describing the interaction between the weight matrix and green using a tensor product $\\otimes$ seems very misleading. This tensor product is also used in the upper part as well, which is deeply misleading, and instead should be expressed as a transposed matrix product. These symbols have a meaning and without redefinition this is pretty confusing.\n3. Figure 1 -> Why is the task optimized framework placed together with the response optimized framework without any clarification of which is which? \n4. Figure 1 -> Why is the dense captioning output also part of the response optimized framework with a rotation equivariant network?\n5. Where are the dense captions coming from? `Line 210` says `An image of size 424 ∗ 424 is divided into grids of size 53 ∗ 53`, but does not otherwise clarify the origin of the captions anywhere in the paper. What model is used here?\n6. `Line 224` please avoid vector matrix products. This assumes row vectors which is not standard.\n7. `Line 237` what are the shapes of the output of function $V_c(x,y)$? Could you describe this sampling in more detail?\n8. Spatial transformer section, `Line 269` it is unclear what $\\theta_2$ is. This is a really important part of the paper and a key claimed contribution. Could the authors mathematically clarify how $\\theta_2$ plays a role?\n9. `Line 286`, what is `AT`?\n10. `Line 297` are the models not voxel wise models?\n11. Figure 2A, the bottom descriptions of the models is very confusing. How is the \"Language Encoder\" being used with \"Semantic Spatial Transformer Readouts\"? \n12. Figure 2B, please use a proper categorical color map for discrete data.\n13. Minor -- Figure 3B, extra space before optimized\n14. Figure 3, how are you defining \"better predicted by Task Optimized\"? Is this the best of ResNet50 or AlexNet? Why use these models when CLIP has been shown to be the best model of visual responses?\n15. All flatmaps have significant wasted space.\n16. Lack of analysis of the spatial transformer networks. While the paper claims STNs as a significant contribution, there is no visualization of the affine parameters for each voxel. Do the affine parameters focus on population receptive fields that are provided in NSD?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "-\tWhy are the models in Figure 2b shown on a continuous color bar?\n-\tWhy were only a subset of the NSD subjects used?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "-\tCompare response optimized and task optimized models directly\n-\tCompared many different model-brain mapping functions\n- Present a new metric for model-brain mapping" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper compares task-optimized vision and language models to brain response-optimized networks on the natural scenes fMRI dataset (NSD), and introduce a new readout mechanism for model-brain mapping based on spatial receptive field and semantic content. They find that response-optimized networks provide the best match to early visual regions, while task optimized vision-language models better match high-level visual regions. They also find their readout mechanism using spatial transformers improves model to brain mapping (though only marginally).\n\nThe comparison between response and task optimized models is interesting, but overall the results provide only a marginal advance in our understanding and computational models of visual cortex. The spatial transformer network readout is novel, but it is not entirely clear what the value of this contribution is. It provides slightly better performance over other methods, but is much more complex (involves integrating an additional Resnet-50 module to weight the channels of each model) and provides only minimal quantitative gains over much simpler methods." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "-\tThe models tested varied along many factors making it difficult to draw strong conclusions about the role of response vs. task-optimization or vision vs. language in model’s performance. For response vs. task, these points could have been made more compelling by training the same architecture on both task and neural responses\n-\tThe biggest issue is that the major findings of this paper have been shown previously (also on the same dataset). Prior work with vision and language models (e.g., Doerig et al.) that showed semantic content is more important for high than low-level visual regions, and Khosla & Wehbe which introduced the response optimized model used here and already showed it predicted NSD responses.\n-\tThe utility of the STN was not well motivated. Figure 2 suggests that the different readout mechanisms provide largely similar results with only minor quantitative differences. This slight boost is unsurprising given how much more complex the spatial transformer network is compared to other readouts." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Why did the authors only use 4 of the 8 participants from NSD? \n\nFigure 1A is confusing. I don’t follow how each of the different readout methods are shown here. Better labels would be very helpful." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "Literature review is comprehensive, and overall, the paper is clearly written. \n\nThe paper is not highly original building on prior readout methods, and recent work conducting large-scale benchmarking of AI models against the brain. However, the addition of dense image captions to extract representations from the language models is a nice contribution to the literature on LM alignment with visual cortex. I have some reservations mentioned below that are impacting my score, but if addressed, I think the paper may constitute a meaningful contribution to the literature." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors aim to evaluate the extent to which LLMs (based on single or dense image captions) predict activity in high-level visual cortex relative to ImageNet-pretrained vision models or neural-response optimized vision models. They introduce a novel readout method that shows higher performance in predicting neural responses relative to linear regression and two other readout methods. They find three distinct regions of visual cortex that are better predicted by vision models, LLMs based on dense captioned images, or LLMs for single image captions." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The paper needs to better justify why a different readout method is necessary. The authors state that the predominant readout method is linear ridge regression, which has high computational and memory demands, but representational similarity analysis (RSA) is nearly as commonly used in the human literature and is less computationally intensive (Kriegeskorte et al, 2008). More importantly, however, the reason that the NeuroAI field tends to rely on linear regression as a readout is based on the logic that we are interested in evaluating the similarity of the representations up to a linear transformation (in representation space) without introducing non-linearities in the readout method. The authors should provide better justification for why a novel readout method is needed within that framework. \n\nThe Semantic Spatial Transformer has greater improvement relative to Ridge regression for the vision model than the language model (Figure 2), and vision models are found to better predict more voxels in high-level visual cortex using the Semantic Spatial Transformer readout (Figure 4) than when using Ridge regression (Figure 5). To me, it is a problem that the readout method does not provide uniform improvements across models. This suggests to me that the readout method is introducing a bias in the conclusions. However, I welcome rebuttal on why this logic is faulty. \n\nFigure 4D shows three regions that respond more to vision model, single captions language model, or dense caption language model, but this is binarizing the difference between each pair of models. However, the claims of these three distinct regions would be strengthened by showing that high-level visual voxels, for example, have additional explained variance by the single caption language model after accounting for the variance explained by vision and dense caption models." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We present a novel Spatial Transformer readout method that enhances accuracy (3-23%), identify 3 brain regions responsive to varying information content; and analyze various neural network models to evaluate their performance several brain regions." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024modeling,\ntitle={Modeling the Human Visual System: Comparative Insights from Response-Optimized and Task-Optimized Vision Models, Language Models, and different Readout Mechanisms},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=12B3jBTL0V},\nnote={under review}\n}" }, "abstract": { "value": "Over the past decade, predictive modeling of neural responses in the primate visual system has advanced significantly, largely driven by various deep neural network approaches. These include models optimized directly for visual recognition, cross-modal alignment through contrastive objectives, neural response prediction from scratch, and large language model embeddings. Likewise, different readout mechanisms—ranging from fully linear to spatial-feature factorized methods—have been explored for mapping network activations to neural responses. Despite the diversity of these approaches, it remains unclear which method performs best across different visual regions. In this study, we systematically compare these approaches for modeling the human visual system and investigate alternative strategies to improve response predictions. Our findings reveal that for early to mid-level visual areas, response-optimized models with visual inputs offer superior prediction accuracy, while for higher visual regions, embeddings from Large Language Models (LLMs) based on detailed contextual descriptions of images and task optimized models pretrained on large vision datasets provide the best fit. Through comparative analysis of these modeling approaches, we identified three distinct regions in the visual cortex: one sensitive primarily to perceptual features of the input that are not captured by linguistic descriptions, another attuned to fine-grained visual details representing semantic information, and a third responsive to abstract, global meanings aligned with linguistic content. We also highlight the critical role of readout mechanisms, proposing a novel scheme that modulates receptive fields and feature maps based on semantic content, resulting in an accuracy boost of 3-23\\% over existing SOTAs for all models and brain regions. Together, these findings offer key insights into building more precise models of the visual system." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Neuro AI", "vision", "deep neural networks", "representations", "fMRI encoding" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/4eb3563f0a891e67101a29eda4a56ba2099e55af.pdf" }, "presentation": null, "primary_area": { "value": "applications to neuroscience & cognitive science" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/35dfabd8266a7daedc718bfe74b77f52e23be8a9.zip" }, "title": { "value": "Modeling the Human Visual System: Comparative Insights from Response-Optimized and Task-Optimized Vision Models, Language Models, and different Readout Mechanisms" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "In Table 1, are those 3 results attained from the same weight parameters and clipping range? Otherwise, are the clipping ranges of the results different?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- The paper highlights the problem of previous work, OCS.\n- The paper shows performance gain compared to OCS.\n- Experimental design and multifaceted analysis of the proposal are commendable.\n- Under the situations that the paper assumes (e.g., the target hardware and bit settings are fixed), significant performance improvements are expected with additional computation overhead." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "Outliers while quantization are one of the representative elements that harm the performance of neural networks.\nWhat makes things worse is that target hardware is usually fixed and hard to change settings such as bit precision.\nThus, if the outlier problem becomes severe and the performance of the quantized model deteriorates, it may be difficult to resolve the problem by using a higher bit.\n\nThere already exists a prior work that tries to resolve this problem which is called OCS.\nBy halving the activation values of outlier channels and duplicating those channels as additional channels, OCS successfully alleviates the outlier problem.\nHowever, OCS causes another problem; rounding error of inliers.\n\nTo mitigate both problems (clipping error of outliers and rounding error of inliers) simultaneously, the paper proposes OCS+.\nThe paper adopts translation instead of halving operation.\nBy doing so, it achieves the same functional effect equivalent to using one more bit with moderate computational overhead.\nWith various experiments, the paper shows that OCS+ outperforms other previous works, even with the same number of channels added by OCS+." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Several problems that OCS already has can be the same problems of OCS+.\n - The computational overhead due to additional channels \n - The purpose of quantization is to run a large model on limited resources. Therefore, additional computational overhead induced by OCS+ has a worse impact on hardware that is hard to adjust bit precision, which is the target of OCS+.\n- The proportion of channels suffering the outlier problem and outlier channel ID can differ according to inputs. Analysis of channel sensitivity with different inputs can be a good experiment." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1.The proposed OCS+ preserves the outlier activations without sacrificing the precision of regular inliers, which allows for a theoretical increase in presentational power from b-bits to (b+1)-bits under the same hardware constraints.\n2.OCS+ is based on the offline mathematical transformations, which does not require additional training or hardware re-design. \n3.Experimental results show OCS+ achieves an improvement over the previous methods both on CNN and ViTs." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a post-training quantization method named OCS+, which aimed at saving outlier activation without affecting the regular inliers. Based on the offline transformation with weight and activation, it does not require additional training. Experimental results show that the proposed method works for both CNN and ViTs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.The presentation need to be improved, such as Page 4.\n2.The actual speed-up need to be evaluated since OCS+ introduces additional computational costs.\n3.Some typos: such as \nLine 93 and Line 99 OCS+-.\nLine 36 quant parameters\nLine 253 translate down?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "I am wondering what the point of extremely low-bit quantization is, such as 2-bit. Does this extremely low-bit make any practical use? Could the author provide some insight into this?\n\nI currently turn to reject this paper for its novelty and experiment comparison." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper provided comprehensive experimental results. \n\n2. This paper has clear logic that makes their method easy to understand." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduced OCS+, a PTQ for solving the outlier in the quantization process. First, this paper demonstrated that outliers are non-trivial. Motivated by OCS, the original version of this paper, OCS+ is introduced. OCS+ duplicates the important (outlier) activation channel and corresponding weight. Thus, OCS+ achieves high performance with the costs of more computation." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Lack of necessary fair comparison. Actually, OCS+ introduces higher computation costs since it increases the weight and activations by 50%. It added 50% extra computation. This makes OCS+ less attractive since the compared methods do not incur extra computation. Also, some related descriptions such as the analysis of FLOP (compared with previous methods)are lacking in the paper.\n\n2. How to select the important activation channels missed in the paper?\n\n3. The implementation details of Table 7 are missing. What are the platform and running software?\n\n4. Running time comparison with previous methods that do not incur extra computation such as BRECQ?\n\n5. According to the paper, the important activation channels seem to be selected on the fly. Does this operation incur more inference time costs?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024ocs,\ntitle={{OCS}+: Improving {PTQ} with Outlier Translation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=12gMsxpu4G},\nnote={under review}\n}" }, "abstract": { "value": "Post-training quantization (PTQ) is an effective technique for accelerating DNN model inference, where activations typically follow a bell-shaped distribution. Since commodity hardware employs a linear quantization grid and limited quantization levels, prior PTQs optimize a clipping threshold to minimize overall quantization error, which excludes outliers from the bell-shaped data. However, outliers are non-trivial for low-bit and lightweight models. Thus OCS (Zhao et al.,2019) proposed to save outliers by halving and duplicating. However, in activation quantization, the original OCS sacrifices the precision of the regular inliers, leading to severe accuracy degradation. To address this, we propose OCS+ to save outlier activation without affecting the regular inliers. Consequently, OCS+ theoretically achieves one-bit higher representation under the predefined bitwidth hardware. OCS+ is based on offline mathematical transformation, thus it does not require additional training or re-design works on hardware. Experiments over CNNs and ViTs demonstrate OCS+ significantly outperforms OCS and help improve current PTQ SOTAs, e.g., OCS+ improves the current SOTAs by 12.73\\% in Acc@1 for W2A2 MobileNet-v2. The code will be released." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Post Training Quantization" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/7d125a4597d2307e999bab894df2a745e250865f.pdf" }, "presentation": null, "primary_area": { "value": "optimization" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "OCS+: Improving PTQ with Outlier Translation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": { "value": "We thank the reviewers for their work." }, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": { "value": "I have read and agree with the venue's withdrawal policy on behalf of myself and my co-authors." } }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. It is recommended that the authors create an adversarial mask to test the attack effectiveness of DAFR in the real world.\n2. It is suggested that the authors add an ablation study of 𝐻: conduct an experiment comparing performance with and without optimizing over multiple images of the attacker.\n3. Does text prompt have an effect on the results?\n4. This paper should include a framework diagram to illustrate the DAFR method, which would enhance readability." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "1. This paper addresses an important issue, namely the security of face recognition models, which is meaningful for both academia and industry.\n2. The paper leverages the ability of diffusion models to generate realistic and natural images to create adversarial face masks, which is an meaningful design." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper focuses on adversarial attacks on face recognition models using face masks, claiming that existing methods lack in attack stealthiness. It proposes a diffusion-based adversarial face mask generation method, titled DAFR. DAFR controls the generation of the diffusion model using adversarial guidance. Additionally, this paper builds a benchmark to evaluate the performance of existing adversarial accessories." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Lack of real-world attack experiments. This paper focuses on the stealthiness of adversarial attacks; however, discussing stealthiness without considering real-world attack implementation seems meaningless. For instance, digital adversarial attack methods can use subtle perturbations undetectable to the human eye, achieving stealthy attacks. The reason existing methods lack stealthiness is due to the need for higher perturbation intensity to achieve real-world attacks. Discussing attack stealthiness without addressing real-world implementation is thus unconvincing. \n2. Evaluation of stealthiness. Stealthiness is a subjective evaluation dimension, and it is unclear if the quantitative metric CMMD used in this paper matches human perception. I suggest adding a user study to support the paper's claims of improved stealthiness.\n3. Lack of ablation experiments. There is insufficient analysis of the effectiveness of key components. For example, to improve robustness, the authors optimize the adversarial pattern on a set of images of the attacker, 𝐻; however, it remains unclear if this design actually improves robustness.\n4. Missing essential references: Towards Effective Adversarial Textured 3D Meshes on Physical Face Recognition, CVPR 2023" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "Yes, Privacy, security and safety" ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weaknesses section." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "It is an interesting work on a relevant problem. The method performs well over previous baselines and including 3D rendering bridges the gap towards real world applicability." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes a method to generate adversarial face masks that can fool face recognition systems in a white box setting. They specifically borrow the adversarial attack framework from AdvDiff." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The attack is a simple adaptation of previous approaches i.e. AdvDiff. \n2. What is R in the Algorithm 1? Should it be $h_i$ instead of $h_n$? \n3. The generation process is firstly done in a white-box setting, while this in itself is not problematic the authors have not included any results on transferability to see whether it can be used for another model? \n4. The attack is not robust against state-of-the-art facial recognition models such as ArcFace. The stealth to attack success rate trade-off for R100 is quite large. SASMask achieves almost the same T-CMMD and M-CMMD scores with significantly higher SR. \n5. Line 208: Which dataset was the ArcFace model trained on? \n6. The test set consisting of 300 images is not statistically significant. \n7. A white mask has a SR 1000 of 0.7083 against the fine-tuned F100 model. Is this scenario even worth studying? This simply means that the model does not perform well to begin with. \n8. Is the attack success impacted by the facial pose?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "How does DAFR perform in the physical world (3D robustness) and against black-box models (attack transferability)?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Applying text-to-image diffusion models to optimizing adversarial accessories is new.\n2. Experiments are conducted on multiple datasets, models, text-guided styles, and with various metrics.\n3. The proposed attack outperforms the baselines regarding stealthiness. \n4. The attempt to form a benchmark is promising." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes to use diffusion models to generate adversarial accessories (a mask) for attacking facial recognition. Two important attack properties, i.e., robustness (resilient to changes in viewing angles and environmental conditions) and stealthiness (do not attract suspicion by, for example, incorporating obvious facial features), are considered." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- This paper has claimed to improve the attack in both robustness and stealthiness. However, the evaluations in the main body are limited to the stealthiness (and the attack strength in the common 2D digital setting). Although some robustness results are added to the Appendix (Table 6), those results show that the proposed attack is worse than the baselines. In addition, no physical, 3D experiments are considered. \n- As acknowledged by the authors, the idea of using diffusion models to generate adversarial examples is not new. However, it is not clearly stated what the specific (technical) challenge is for generating adversarial masks, compared to other forms of perturbations, such as the makeups. Without identifying such challenges, the technical novelty of this paper is not clear.\n- It is interesting to report the results for different text prompts. However, since different prompts lead to dramatically different attack performance (see Table 4), it would be necessary to understand the relation between the prompt and the performance and finally learn to select the optimal one.\n- Presentation issues:\nThis paper contains lots of technical details but lacks an overview of the whole attack pipeline (maybe in the form of a flowchart).\nBefore introducing their method, the paper should include foundational knowledge related to adversarial masks and clarify the meanings of the mathematical symbols. For instance, the meanings of x and I in Algorithm 1 are not defined." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "Please see the weaknesses above." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The literature review is thorough and effectively encompasses key works relevant to the field." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose Diffusion Attack against Facial Recognition (DAFR), a method capable of generating robust and stealthy face masks for dodging recognition systems using the guidance of textual prompts to determine the style. A new benchmark is also presented, the Face Accessory Attack Benchmark (FAAB) which includes a set of standardized tests and procedures to evaluate the performance of accessories." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Motivation: While I recognize the significance of prior works highlighting face masks as a potential threat vector during the COVID-19 pandemic, their prevalence has markedly declined since then. As a result, I find it challenging to accept the notion that face masks, irrespective of printed patterns, can currently be considered a genuinely stealthy accessory.\n2. Writing quality: the submission is mainly held back by the writing quality and lack of clarity in most sections. These are mainly focused around the method section (Section 2), which seems a bit unprofessional, lacking formulaic descriptions of relevant preliminaries and the attack’s full pipeline (e.g., how the components interact with each other) and overall extremely unorganized. I suggest the authors reorganize (e.g., split the section into subsections, each describing the main steps of the attack) and improve (formalize the entire pipeline) this section for better clarity of the novelties they propose.\n3. Novelty: The proposed method appears to lack substantial originality, as it primarily combines existing approaches adapted to this specific task, without introducing significant additional contributions. For instance, the mask rendering technique for facial images is adopted from Zolfi et al. (2022), while the diffusion-based adversarial attack approach is drawn from Dai et al. (2023), with adjustments made for the face recognition domain. Although the authors mention the use of textual prompts to control style, the method section lacks a clear methodological explanation of this aspect, including details on how prompts are selected and their impact on the attack's objectives.\n4. Evaluation: While the authors assess their attack against various baselines and across multiple models and datasets, several critical aspects remain unaddressed:\n- Shallow Analysis – The authors predominantly present empirical results without delving into deeper insights, examining edge cases, or discussing unexpected findings. \n- Missing Ablation Studies – For instance, Equation 1 claims that a scaling function is superior to a static value, yet no comparative analysis is provided to validate this assertion. \n- Lack of Transferability Experiments – A crucial aspect of adversarial attacks is their transferability to models beyond those on which they were trained. Testing transferability could offer valuable insights into the practicality of the proposed attack. \n- Absence of Real-World Experiments – Although digital evaluations form the core of the paper, an attack using a practical accessory would benefit from real-world testing to demonstrate efficacy beyond digital scenarios. \n- Implementation Details – The section includes an excess of low-level details (specific steps for each decision), which detracts from the key information. I recommend prioritizing content between the main paper and the appendix, allowing for more space for additional experiments, such as those in Sections D, E, and F of the appendix. \n- Results – The results across most models do not consistently outperform the baselines (notably SASMask). Ideally, the proposed method should exhibit at least comparable performance to baselines while enhancing stealthiness. The current configuration seems to prioritize stealthiness at the expense of attack success.\n\nMinor comments:\n1. Line 89 – CMMD is not introduced until line 446 (not even a reference to it).\n2. Line 125 – unclear why $f$ is mentioned here.\n3. Line 137 – unclear what $R$ is.\n4. Line 301 – table 3 is too far from where it was mentioned in the text. Maybe the table could be split to better fit in the flow of the text." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We propose a diffusion based adversarial face mask attack, and a benchmark for evaluating face mask attacks." }, "_bibtex": { "value": "@misc{\nlewis2024text,\ntitle={Text To Stealthy Adversarial Face Masks},\nauthor={Ben Lewis and Thomas Moyse and James Parkinson and Elizabeth Telford and Callum Whitfield and Ranko Lazic},\nyear={2024},\nurl={https://openreview.net/forum?id=12iSWNLDzj}\n}" }, "abstract": { "value": "Recent studies have demonstrated that modern facial recognition systems, which are based on deep neural networks, are vulnerable to adversarial attacks, including the use of accessories, makeup patterns, or precision lighting. However, developing attacks that are both robust (resilient to changes in viewing angles and environmental conditions) and stealthy (do not attract suspicion by, for example, incorporating obvious facial features) remains a significant challenge. In this context, we introduce a novel diffusion-based method (DAFR) capable of generating robust and stealthy face masks for dodging recognition systems (where the system fails to identify the attacker). Specifically our approach is capable of producing high-fidelity printable textures using the guidance of textual prompts to determine the style. This method can also be adapted for impersonation purposes, where the system misidentifies the attacker as a specific other individual. Finally, we address a gap in the existing literature by presenting a comprehensive benchmark (FAAB) for evaluating adversarial accessories in three dimensions, assessing their robustness and stealthiness." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": { "value": [ "~Ben_Lewis1", "~Thomas_Moyse1", "~James_Parkinson1", "~Elizabeth_Telford1", "~Callum_Whitfield1", "~Ranko_Lazic1" ] }, "authors": { "value": [ "Ben Lewis", "Thomas Moyse", "James Parkinson", "Elizabeth Telford", "Callum Whitfield", "Ranko Lazic" ] }, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "facial recognition", "adversarial accessories", "diffusion models", "adversarial benchmarks" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": { "value": "lewis|text_to_stealthy_adversarial_face_masks" }, "pdf": { "value": "/pdf/4433a437b437976e7c40bdd6156a2df66dd7f0eb.pdf" }, "presentation": null, "primary_area": { "value": "alignment, fairness, safety, privacy, and societal considerations" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/1e755401d06d1021e2d8e8fa0f3b8e040ba8d2e9.zip" }, "title": { "value": "Text To Stealthy Adversarial Face Masks" }, "venue": { "value": "ICLR 2025 Conference Withdrawn Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Withdrawn_Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "I don't quite understand how the 2D feature map is constructed. It is said that it directly comes from the neural network model, but how? Could the authors provide more details?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. By using Voronoi tessellation and introducing a ridge confidence metric, the paper provides a more detailed and interpretable visualization of decision boundaries, offering insights into the complexity and uncertainty of boundary regions.\n1. The authors compare multiple AL sampling methods (e.g., entropy, margin, BALD dropout, KMeans) on distinct datasets, providing valuable insights into the behaviors and trade-offs of each approach in different scenarios.\n1. The visualization effectively demonstrates how models handle uncertainty due to limited training samples and noisy regions, which is beneficial for identifying optimal query strategies for different types of uncertainties.\n1. While focused on active learning, the proposed visualization technique has potential utility in other areas of machine learning that require understanding complex decision boundaries in high-dimensional spaces." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces a Voronoi tessellation-based visualization method for active learning (AL) to address the limitations of traditional visualizations, which often lack detail in uncertain or sparse boundary areas. By using a ridge confidence metric with Voronoi cells, the approach offers a clearer, more informative view of decision boundaries, aiding in analyzing different AL sampling strategies. Tested on MNIST and CIFAR-10 datasets, the method reveals unique sampling behaviors across entropy, margin, and diversity-based strategies, helping clarify how each handles uncertainty and impacts model learning." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. This paper lacks a review of related work. It's important to examine the literature in the related fields, especially other visualization methods developed for active learning.\n1. While the work in the paper is valuable, it also lacks of the baseline methods. The paper demonstrates the effectiveness of Voronoi tessellation and ridge confidence in active learning throught several case studies, but it does not prove how much better it is compared with other visualization methods. I suggest a quantitative analysis with different visualization methods, like a user study.\n1. Section 3 is over lengthy. The authors should consider breaking it down. Subsection 3.1 and Subsection 3.2 should be independent sections.\n1. The discussion of the 2D feature map is missing. How much does space distortion affect the Voronoi tesselletion construction? The neighbors in the original space are not always the same in the feature space. How reliable is the decision boundary in this case? Since everything is visualized in the feature map, it's essential to give a comprehensive discussion on the feature map itself.\n1. The visualization strategies are informative, but the implementation in the plots are really hard to follow. I have several suggestions for improvement:\n - Figure 2, 3, 5, 6, 7, and 8 are visualization results, but they are too small and dense to read, which is fatal for a visualization paper. At least the authors should put high resolution images in the appendix.\n - In figure 5, the colormap of confidence interval should be different from that of the classes. I suggest different levels of gray to show the confidence interval\n - In the visualization plots, the scatter points, Voronoi ridges, and cells are crowded and overlapping, causing severe visual clutter. Actually, it's not necessary to show all the information in a single plot at once. It only makes the plot massive. One way to display a large quantity of information is to enable user interactions. For example, enable users to choose classes they are interested in, add a slider to show different levels of confidence, brush to zoom in on a decision boundary. \n - The Error Detection visualization results are interesting. It gives more information on how the active learning model behaves. I suggest putting multiple training iterations of Error Detection plots in the appendix to visualize its evolution over time. \n\nWhile I believe this work makes good contribution, I'm afraid the authors cannot resolve my concerns in a reasonable time. Therefore, I recommend a weak reject." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "- It is unclear why one can assess the confidence of the predicted ridges with the representative Voronoi center points $\\mathbf{p_i}$ and $\\mathbf{p_j}$ (equation 2). Although each ridge can be represented by its representative point $\\mathbf{p}$, the predicted probabilities close to the decision boundary may differ strongly from their center. Could you please clarify why using the point $\\mathbf{p}$ gives an accurate prediction of the confidence?\n- How did you decide on which snapshots to visualize?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The paper sheds light on the crucial topic of getting insights into machine learning algorithms and provides visual results that highlight apparent differences between sampling strategies.\n- The chosen approach of this paper is generalizable to various machine learning models performing prediction tasks.\n- The authors provide extensive visualizations for each sampling strategy and compare their characteristics with the help of side-by-side visualizations.\n- The paper examines a high number of sampling strategies." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The proposed work deals with the challenging issue of getting insights into sampling strategies of active learning. Addressing this issue, the authors developed a novel visualization approach: The feature space of the samples is projected to a 2D plane and segmented into Voronoi cells. Ridges of the Voronoi cells are used to construct the decision boundary. Additionally, the confidence of this decision boundary is assessed by leveraging predicted probabilities of samples for each ridge. Thus, the confidence of the decision boundary varies locally. The authors apply the confidence decision boundary to visualize the learning and sampling behavior of eight active learning strategies on MNIST and CIFAR-10. Using various other visualizations, they qualitatively depict the characteristics of the sampling strategies and conclude that uncertainty caused by insufficient training samples may be effectively tackled by sampling. In contrast, uncertainty in noisy regions may be hard to tackle." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The paper contains errors in writing. For example, the sentence \"Since points on either side of the predicted ridges belong to different predicted classes, the confidence of the predictions vary, different sections of the decision boundary carry varying degrees of informative value due to differences in prediction confidence.\" in lines 110-113 should be corrected. \n- Visualizations lack proper color encoding. For example, the decision boundary, a continuous value, is represented with a qualitative color scale (see Figure 5).\n- It remains unclear which snapshots of the feature space were taken for visual elaboration. For example, it is not described why the third round of training was used to inspect the decision boundary in Figure 5. Thus, the authors' conclusions are only based on single visual findings, making them hard to retrace.\n- A clear methodology for selecting snapshots for inspection would improve the clarity and usefulness of the authors' approach. Furthermore, the evaluation relies on observations of a single training run - a more extensive evaluation with a higher number of training runs per sampling strategy would be a good starting point.\n- Some visualizations lack proper annotations, making them hard to understand. Figure 4 especially needs annotations on what the different colors encode. Similarly, which training rounds do we see in Figure 2?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "One aspect I am particularly concerned about is the scalability of this visualization. Currently, the data volume is relatively small, but with a large number of categories, how can the scalability of this visualization be ensured? I would suggest that the authors provide more discussion and experimental results on this point. When the dataset is large, the entire Voronoi becomes complex, with very small cells and densely packed ridges that obscure each other, making it visually unfriendly. I doubt whether users other than authors can derive similar insights from such dense and non-interactive visualizations. The authors could provide quantitative and qualitative user surveys to demonstrate the effectiveness and usability of their method. The authors could consider further optimizations, such as clustering before segmentation or refining the segmentation rules, rather than just using Voronoi tessellation.\nThe visualization method seems to lack interactivity. Providing an interactive tool would enhance the appeal of this work. Additionally, the authors do not explicitly mention which dimensionality reduction method was used. Different dimensionality reduction techniques may affect subsequent Voronoi and decision boundary construction. The authors should provide explanations and evaluations of this aspect." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "I’m glad to see that some work at the ICLR conference focuses on data visualization. The insights provided by data visualization can better help users understand some of the underlying aspects behind the models. Additionally, visualization on the existing dataset is able to provide better insights and differentiation capabilities." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces the confidence decision boundary visualization method, which provides an interpretable comparison of various active learning strategies, yielding valuable insights. The experimental design and discussion are thorough.\nI’m glad to see that some work at the ICLR conference focuses on data visualization. The insights provided by data visualization can better help users understand some of the underlying aspects behind the models. Additionally, visualization on the existing dataset is able to provide better insights and differentiation capabilities.\n\nOne aspect I am particularly concerned about is the scalability of this visualization. Currently, the data volume is relatively small, but with a large number of categories, how can the scalability of this visualization be ensured? I would suggest that the authors provide more discussion and experimental results on this point. When the dataset is large, the entire Voronoi becomes complex, with very small cells and densely packed ridges that obscure each other, making it visually unfriendly. I doubt whether users other than authors can derive similar insights from such dense and non-interactive visualizations. The authors could provide quantitative and qualitative user surveys to demonstrate the effectiveness and usability of their method. The authors could consider further optimizations, such as clustering before segmentation or refining the segmentation rules, rather than just using Voronoi tessellation.\nThe visualization method seems to lack interactivity. Providing an interactive tool would enhance the appeal of this work. Additionally, the authors do not explicitly mention which dimensionality reduction method was used. Different dimensionality reduction techniques may affect subsequent Voronoi and decision boundary construction. The authors should provide explanations and evaluations of this aspect." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "One aspect I am particularly concerned about is the scalability of this visualization. Currently, the data volume is relatively small, but with a large number of categories, how can the scalability of this visualization be ensured? I would suggest that the authors provide more discussion and experimental results on this point." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Decision Boundary Representation: How do the fixed Voronoi tessellation results relate to the updated feature vectors during training? Can you explain how this approach ensures an accurate representation of the model's decision boundaries throughout the training process?\n2. Generalizability: How does the method work on more complicated datasets, such as ImageNet (with more classes) and CUB-200-2011 (fine-grained classification tasks)\n2. Boundary Effectiveness: Can you elaborate on how the Voronoi tessellation approach adds value beyond insights obtainable from scatterplots? What specific contributions does it provide in terms of understanding boundary effectiveness that traditional methods do not?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Comprehensive Comparison of Active Learning Strategies: The evaluation includes a systematic comparison of various active learning strategies. These potentially contribute to the broader discourse on active learning methods and offer practical guidance for future research in this area.\n2. Effective Use of Visualization Techniques: The paper incorporates visualization methods to facilitate the analysis of data and models. This makes the analysis and findings more accessible and intuitive." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents a Voronoi tessellation-based method to visualize decision boundaries in active learning. By highlighting ridges between cells with differing predictions, this method provides a clearer view of the decision boundaries. In addition, it introduces a ridge confidence metric to quantify prediction uncertainty. Experiments on MNIST and CIFAR-10 illustrate the effectiveness of this approach for analyzing various active learning strategies, providing insights into the behavior and effectiveness of different sampling techniques." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. **Concerns about Decision Boundary Generation**. From the figures, the projection results and the Voronoi tesselation results appear to be fixed across multiple rounds. However, as feature vectors update during training, the projection results and the decision boundaries should also be updated. It is essential to clarify how well these fixed tessellations with the updated predictions capture the model's behavior. In addition, the use of 2D projections to represent high-dimensional decision boundaries raises concerns, as results can vary significantly based on the selected projection method and parameters.\n2. **Insufficient evaluation**. While this paper compares different active learning strategies and summarizes some insights, this evaluation is not sufficient and rigorous. On the one hand, the evaluation is only conducted on MNIST and CIFAR-10, which only contain a few classes with substantial differences between them. It remains uncertain how the proposed method performs with more classes or finer distinctions. On the other hand, the evaluation of boundary effectiveness is inadequate. Many insights, such as oversampling on the boundary region in the uncertainty-based methods, can be identified from scatterplots alone, making the Voronoi approach seem unnecessary.\n3. **Omission of Relevant Literature**. The paper overlooks significant works that utilize Voronoi tessellation for visualizing sample boundaries. For instance, Chen et al. [*] use it to visualize the samples and boundaries in the analysis of label propagation. Including such references would enhance the contextual foundation of the research.\n[*] Interactive Graph Construction for Graph-Based Semi-Supervised Learning. TVCG 2021." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024voronoi,\ntitle={Voronoi Tessellation-based Confidence Decision Boundary Visualization to Enhance Understanding of Active Learning},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=13G5KXm98a},\nnote={under review}\n}" }, "abstract": { "value": "The current visualizations used in active learning are quite basic, making it difficult for researchers to effectively observe and analyze the practical performance of different sampling strategies. To address this issue, we introduce a more informative visual evaluation approach observation metric, the confidence decision boundary, which is generated through Voronoi tessellation and evaluated using ridge confidence, a newly proposed measure. This approach enhances the information content in boundary regions where data distribution is sparse. Based on the confidence decision boundary, we conducted a series of visualizations to evaluate various active learning query strategies. These visualizations are able to capture nuanced variations regarding how models based on different strategies perform sampling, the characteristics of points selected by various methods, and the impact of newly sampled points on the model. This enables a much deeper understanding of the underlying mechanisms of existing query strategies." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Decision Boundary", "Visualization", "Active Learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/750db6dff0610ce08dd03633f85849e0c286beeb.pdf" }, "presentation": null, "primary_area": { "value": "other topics in machine learning (i.e., none of the above)" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Voronoi Tessellation-based Confidence Decision Boundary Visualization to Enhance Understanding of Active Learning" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Could you elaborate on the choice of baseline models? How does EEGMamba’s performance compare with multi-task EEG models specifically designed for generalization across diverse tasks?\n\nHow does the ST-Adaptive module capture interactions between channels and temporal features, which are critical in EEG data? Have you considered modeling these dependencies more explicitly?\n\nHow would EEGMamba perform on out-of-distribution tasks or tasks not seen during training? Have you tested its ability to generalize to entirely new EEG task types?\n\nCould you clarify how this work builds on or differs from previous applications of SSMs or Mamba models in EEG classification? Adding this context could help readers better understand the specific contributions of EEGMamba.\n\nCan you evaluate this on longer contexts to better get a sense for the necessity and usefulness for Mamba?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "EEGMamba achieves state-of-the-art performance across multiple EEG tasks, demonstrating robust multi-task capability. Its bidirectional Mamba blocks enable efficient handling of long sequences, avoiding the high memory demands of Transformer models. The flexible ST-Adaptive module supports EEG signals of varied lengths and channels, and the task-aware Mixture of Experts (MoE) enhances task-specific accuracy, reducing interference across tasks. This adaptability and strong generalization across datasets position EEGMamba as a versatile model for EEG classification." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents EEGMamba, a model tailored for multi-task EEG classification. It aims to overcome the limitations of existing models in terms of computational complexity and generalization across tasks with varying signal lengths and channel counts. EEGMamba integrates three main innovations: the Spatio-Temporal-Adaptive (ST-Adaptive) module for unified feature extraction, Bidirectional Mamba to balance accuracy and computational efficiency, and a Task-aware Mixture of Experts (MoE) to handle the differences and similarities across EEG tasks. Evaluated across eight public datasets and covering seizure detection, emotion recognition, sleep stage classification, and motor imagery, EEGMamba demonstrates strong performance and adaptability." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "EEGMamba’s evaluation lacks comparison the newest baseline models (notably LaBRaM from ICLR 2024), which limits the interpretability of its reported gains. The Spatio-Temporal-Adaptive (ST-Adaptive) module, while flexible for varying signal lengths and channel counts, may not adequately capture complex channel-time dependencies crucial in EEG data. Furthermore, training the ST module on a per-task basis could lead to representations that are too specialized, reducing generalizability, particularly for out-of-distribution (OOD) tasks, which were not included in the study. Additionally, the paper lacks a related works section detailing prior applications of SSMs or Mamba in EEG classification, making it difficult to contextualize EEGMamba’s specific advancements in this space." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "In Section 4.2's experimental comparison, was the model trained using all the datasets at once? Could there be interactions between the datasets? Would training the model on each dataset separately improve the performance metrics? Have you conducted any experiments on this? I'm quite interested." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "### **Originality**\nThe originality of EEGMamba lies in its novel approach to EEG classification through the integration of bidirectional Mamba, Spatio-Temporal-Adaptive (ST-Adaptive) modules, and task-aware Mixture of Experts (MoE). The innovative combination of these elements addresses the computational complexity and variability in signal length and channels, which are critical challenges in EEG classification. This creativity, particularly in applying multitask learning to EEG signals, represents a significant advancement in the field.\n\n### **Quality**\nThe quality of this work is demonstrated through rigorous evaluations on multiple publicly available EEG datasets. EEGMamba's superior performance in seizure detection, emotion recognition, sleep quality, and emotion recovery highlights its robustness and effectiveness. The model's ability to handle long sequences and adapt to different feature extraction tasks while maintaining high accuracy and fast inference speed.\n\n### **Clarity**\nThe authors provide a detailed description of the EEGMamba architecture and its components. The step-by-step explanation of how the bidirectional Mamba, ST-Adaptive module, and task-aware MoE are integrated and function together contributes to a well-structured and coherent narrative.\n\n### **Significance**\nThe model's design, which allows for the efficient capture of both task-specific and general features, has the potential to transform how EEG data is processed and analyzed. This can lead to more accurate and comprehensive analyses of complex brain signal data, benefiting various applications such as medical diagnostics and cognitive research." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces EEGMamba, a novel EEG classification network designed for multitask learning. It integrates spatiotemporal adaptive (ST-Adaptive) modules, bidirectional Mamba, and a mixture of experts (MoE) approach. This addresses challenges in EEG classification, such as computational complexity and variations in signal length and channels. The model efficiently handles long sequences and adapts to feature extraction while capturing both task-specific and general features. Evaluations on multiple public EEG datasets demonstrate that EEGMamba outperforms existing models in seizure detection, emotion recognition, sleep quality, and emotion recovery." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Overall, the experiments in this paper are comprehensive; however, in Section 4.1, the discussion on single-channel and multi-channel models only compares memory usage and inference speed, without evaluating the impact of multi-channel models on performance metrics. Additionally, the t-SNE visualization lacks layer-by-layer analysis of the model's influence on clustering results, which does not adequately demonstrate the feature extraction capability of each layer. It is recommended to visualize feature clustering by module." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Plz go and check weakness" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "EEGMamba achieves superior memory efficiency and faster inference speed than traditional transformer-based models, especially on longer EEG sequences, thus proving its practical value. \n\nThe proposed model demonstrates an approach to handling EEG data of varying lengths and channels, incorporating a class token for temporal adaptability and a task-aware MoE to distinguish task-specific features." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper \"EEGMAMBA: Bidirectional State Space Model with Mixture of Experts for EEG Multi-Task Classification\" presents EEGMamba, a multi-task learning framework for EEG classification tasks, addressing challenges related to signal length and channel variability. EEGMamba integrates a Spatio-Temporal-Adaptive (ST-Adaptive) module, bidirectional Mamba blocks, and a task-aware Mixture of Experts (MoE) to enhance adaptability and task-specific processing across diverse EEG datasets. The ST-Adaptive module standardizes data of various lengths and channel numbers, while the bidirectional Mamba captures temporal dependencies in EEG sequences, and the task-aware MoE module selects experts based on task, enhancing classification accuracy and generalization. Tested on eight datasets spanning four task types (seizure detection, emotion recognition, sleep stage classification, and motor imagery), EEGMamba achieves state-of-the-art results, demonstrating superior efficiency, memory usage, and accuracy across tasks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Limited Contribution: The author claims that EEGmamba is the first universal EEG classification network to effectively implement multi-task learning for EEG applications. However, several established methods are available that can be applied to multi-task EEG learning, as cited in [1][2][3].\n\n2. Inappropriate Comparisons: The choice of baselines for comparison lacks relevance. For instance, using AttnSleep [3] as a baseline for seizure detection/emotion recognition is incongruous, as it is specifically designed for sleep staging. Additionally, the author does not include multi-task learning methods as baselines, instead comparing against smaller models tailored to individual tasks. For fair assessment, each specific task should be compared to the most relevant model designed for that purpose. To evaluate cross-task capabilities, comparisons should involve multi-task learning methods like those in [1][2][3][5][6].\n\n3. Efficiency Evaluation: The author should provide quantitative evidence demonstrating the proposed method's efficiency advantages over previous approaches.\n\n\n\n[1] Jiang, W., Zhao, L., & Lu, B. L. Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI. In The Twelfth International Conference on Learning Representations.\n\n[2] Chen, Y., Ren, K., Song, K., Wang, Y., Wang, Y., Li, D., & Qiu, L. (2024). EEGFormer: Towards transferable and interpretable large-scale EEG foundation model. arXiv preprint arXiv:2401.10278.\n\n[3] Jiang, W. B., Wang, Y., Lu, B. L., & Li, D. (2024). NeuroLM: A Universal Multi-task Foundation Model for Bridging the Gap between Language and EEG Signals. arXiv preprint arXiv:2409.00101.\n\n[4] Eldele, E., Chen, Z., Liu, C., Wu, M., Kwoh, C. K., Li, X., & Guan, C. (2021). An attention-based deep learning approach for sleep stage classification with single-channel EEG. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 29, 809-818.\n\n[5] Zhang, D., Yuan, Z., Yang, Y., Chen, J., Wang, J., & Li, Y. (2024). Brant: Foundation model for intracranial neural signal. Advances in Neural Information Processing Systems, 36.\n\n[6] Wang, C., Subramaniam, V., Yaari, A. U., Kreiman, G., Katz, B., Cases, I., & Barbu, A. (2023). BrainBERT: Self-supervised representation learning for intracranial recordings. arXiv preprint arXiv:2302.14367." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Introduction: (line 49-50):\n\nThe authors claim that CNNs are unable to handle long EEG signals, citing three papers: (Sakhavi etal.,2018), (Thuwajit etal.,2021) and (Schirrmeister etal.,2017). However, none of these studies provide evidence to support such a conclusion. In fact, Thuwajit et al. (2021) proposed EEGWavenet, which utilizes a multiscale CNN-based spatiotemporal feature extraction module. This module gradually increases its receptive field to match the length of the input EEG signal, indicating that CNNs can handle the global input sequences in certain contexts. While the authors' claim is not entirely substantiated, it raises interesting questions that I would like to see addressed:\n\na. For EEG classification, what defines a \"long\" signal in terms of sample size? At what point does a short receptive field cause CNN performance to degrade?\n\nb. Is global sequence modelling truly necessary for long-term EEG signals? From a neuroscience perspective, how much does brain activity from, say, 10 seconds ago, affect the current state? Which types of tasks specifically require such long-term modelling?\n\n2. section 2.2 ST-Adaptive Module:\n\nThe authors propose a spatio-temporal-adaptive module that transforms arbitrary EEG inputs into a uniform feature dimension, as depicted in Figure 3. However, I have several concerns:\n\na. Scalability to New Datasets: Is this approach scalable to new datasets once the model is trained? Given that different Conv1d layers are used to transform varying numbers of EEG channels into a common hidden state in the Spatial-Adaptive Convolution, how flexible is this method for new tasks where the number of channels may differ from the training datasets? Even if the number of channels is the same, the channels themselves vary. It appears that the model learns dataset-specific spatial filters rather than a truly global spatial representation, which may limit its generalizability to new datasets, which is a big issue for a backbone model, since people would like to use it on different tasks later on.\n\nb. Tokenize Layer: In the tokenize layer, two branches of CNN are used: one with a short kernel (15) and another with a long kernel (49). Was this choice based on experimental results? Why are there only two branches, and why were kernel sizes 15 and 49 specifically chosen? Since there’s no discussion of this configuration in the ablation study, it’s unclear whether this is the most optimal setup.\n\n3. section 3.2 Data Division:\n\nIn this study, the 5-fold cross-validation experiment was implemented as leave-subject-out, which is not a common approach in the BCI field due to the significant subject variability. This evaluation approach faces two challenges: (1) developing a population model trained on a group of subjects, and (2) addressing subject transfer by evaluating the model on unseen subjects. This raises several concerns:\n\na. Subject Variability Impact: In tasks with minimal subject variability, such as seizure detection and sleep stage detection, the classification results are high, as shown in Figure 1. However, tasks like motor imagery and other BCI tasks exhibit high subject variability, which severely impacts model performance. This is evident in the low accuracies achieved on the BCI-IV-2a task (44%) and the SEED task (57.2%), which are insufficient for practical BCI applications. A discussion on this performance discrepancy is necessary to demonstrate how the proposed model addresses these challenges and whether it shows superiority in domains with high subject variability.\n\nb. Performance Discrepancy with Benchmark Models: Many benchmark models, such as EEG Conformer (Song et al., 2022), were not designed to handle population transfer. EEG Conformer, for instance, was trained in a subject-specific manner and achieved state-of-the-art performance on the same BCI-IV-2a (78.66%) and SEED (95.30%) datasets. However, in this study, the EEG Conformer’s performance dropped significantly to 35.2% and below 50%, respectively. Could the authors explain this stark performance difference? Is it primarily due to the leave-subject-out evaluation setting? If so, would the proposed EEGMamba model also retain its high performance if evaluated in a subject-specific manner like EEG Conformer?\n\nc. Use of Separate Test Sets: Datasets like BCI-IV-2a include a separate test set specifically intended for evaluation. In this study, it is unclear whether the authors utilized this test set. Since many studies report performance on this designated test set, comparing the classification accuracy reported here with those from other studies may not be straightforward. Clarification on whether the official test set was used, or if an alternative test split was applied, is needed to ensure a fair comparison with prior work.\n\n4. section 4.1 Single-Task EEGMamba Performance Comparison\n\nLine 377: The authors mention the memory and inference time challenges of transformer models when handling long sequences. Similar to my previous concern, what qualifies as a \"long\" sequence in terms of signal length for transformers to encounter this bottleneck? Is there a real-world application where such long sequences need to be tackled? Many EEG tasks are only a few seconds in length, so it would be helpful to clarify the practical need for handling significantly longer sequences.\n\n5. section 4.2 EEGMamba for EEG Multi-Task Classification\n\nWhile the idea of training a single model to perform well across multiple datasets is interesting, its practical application is unclear. From Figure 1, the single-task models appear to achieve similar or even better performance on 5 out of the 8 benchmarked datasets (Siena, BCI-IV-2a, Shu, SEED, CHB-MIT). This raises a few questions:\n\na. Benefit of Multi-Task Training: Is there a demonstrable benefit to multi-task training? Specifically, is there any statistical difference between the performance of the single-task model and the multi-task model on the datasets? It would be helpful to clarify whether multi-task training consistently improves performance or if its benefits are marginal.\n\n6. Lines 454-456:\n\nThe authors argue that the model only needs to be trained once. However, the analysis was performed offline on 8 selected public datasets solely. To assess whether this model can be applied in real-world scenarios, the following questions need to be addressed:\n\na. Cross-Session Issues: In practice, many EEG-based applications require short calibration sessions to adjust for cross-session variability in subject-dependent models. Does the proposed model also require such calibration? If calibration is needed, would this involve further training or fine-tuning of the pre-trained model? In the case of the multi-task model, would calibration require data from other tasks as well, or could it be done independently?\n\nb. Generalization to New Datasets: How well does the model perform on a new dataset that belongs to one of the pre-trained tasks? If the model only needs to be trained once, does this mean it can be directly applied to similar tasks without additional training? For example, how would the model handle BCI-IV-2B, which has only 3 channels but is still a motor imagery task, or another sleep stage classification dataset? If yes, how would the model manage inconsistencies in the number of channels, and what performance can be expected? If not, wouldn’t this imply that researchers would still need to retrain the model, making it not different from other models?\n\nb. Domain-Specific Advantages: If multi-task training is advantageous, is this benefit domain-specific? For example, should a researcher developing a motor imagery decoder expect better results from multi-task training? How can researchers determine whether a single-task or multi-task approach will yield better performance for a specific domain or task?\n\nc. Practicality of Multi-Task Training: In practice, most researchers focus on specific tasks and experiments, collecting data for single tasks. Does this multi-task approach suggest that researchers in the future should also record additional tasks or rely on public datasets to improve performance? Or is there a scenario where it would make sense for a model to simultaneously classify motor imagery, emotion, sleep stages, and seizure events? More guidance on when and why to use multi-task training would be valuable.\n\n7. section 4.4 Ablation Study:\n\nIn Figure 6, it appears that the performances of the different configurations, except for the single-directional Mamba, are quite similar. Is there a statistically significant difference between these configurations? It would be helpful to include a discussion on whether the variations in performance are meaningful or simply within the margin of error.\n\n8. Conclusion lines 526-527:\n\nThe authors claim that EEGMamba is the first model to truly implement multi-task learning for EEG applications. However, I am curious about how EEGMamba differs from a simpler approach, such as using a shared backbone model as a feature extractor with separate classification heads for different tasks, as done in [1]. Could the authors clarify the key differences between the proposed model and such an approach using MoE?\n\n[1] Xie Y, Wang K, Meng J, Yue J, Meng L, Yi W, Jung TP, Xu M, Ming D. Cross-dataset transfer learning for motor imagery signal classification via multi-task learning and pre-training. J Neural Eng. 2023 Oct 20;20(5). doi: 10.1088/1741-2552/acfe9c. PMID: 37774694.\n\n9. Conclusion line 531:\n\nThe authors claim that the proposed model can better learn the commonalities among EEG signals from different tasks. However, what specific commonalities are being referred to? Is there any interpretation or evidence to support this claim? It would be helpful to understand how these commonalities are identified and whether the model offers any insight into them.\n\nAddressing the above questions could help clarify specific weaknesses and improve the overall impact of the study, more specifically:\n\nQuestion 5: Answering this question would clarify Weakness Point 1, providing more insight into the motivation for multi-task training.\n\nQuestions 1, 2(a), and 6: Answering these would address Weakness Points 2 and 3, which would significantly enhance the study's potential impact and score by clarifying practical applications and the model’s performance on motor imagery datasets.\n\nQuestions 4 and 7: These would address Weakness Point 4 by supporting the comparisons with a statistical validation." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The authors successfully introduce the new Mamba architecture to EEG decoding, achieving strong results across 8 public datasets.\n\n2. Mamba demonstrates memory efficiency and fast inference, making it advantageous for real-world applications.\n\n3. The model effectively addresses the multi-task classification problem, showcasing the feasibility of training a single model for multiple downstream tasks." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors introduce EEGMamba, a model designed for multi-task EEG classification. EEGMamba consists of an ST-Adaptive module that learns spatial filters for each task, transforming EEG inputs with varying channel counts into a uniform feature space. The module then tokenizes the data using both small and large kernels to capture short-term and long-term features. These tokens are processed by a BiMamba backbone with task-aware Mixture of Experts (MoE) layers, enabling the model to capture both task-specific and shared features. Finally, each task has a dedicated classification head.\n\nEEGMamba allows for multi-task EEG classification in a single training session. The authors evaluated the model’s performance against five other models across eight public datasets, covering tasks such as epilepsy detection, sleep stage classification, emotion recognition, and motor imagery. The experiments used a 5-fold cross-validation approach, where specific subjects were reserved for the test set in each fold. Results show that EEGMamba outperformed competing models under this evaluation, and it demonstrated efficient memory usage and inference speed, particularly with long-sequence data." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The motivation for multi-task training is somewhat unclear. The authors should clarify why multi-task training is necessary and how it can be beneficial for specific applications.\n\n2. Certain questions remain unanswered regarding the practical use of the proposed model, which limits its potential impact. For instance, under the current evaluation scheme, it is unclear how the model would generalize to new datasets or subjects, particularly when new datasets vary in channel count. Does introducing a new dataset require retraining or additional multi-\n\ntask training even if it is single-task? Additionally, what would be the training strategy for developing a subject-dependent model within the multi-task framework if only one task is available from the subject? To address this, the authors could consider testing the model on an additional dataset to evaluate whether the pre-trained model can transfer effectively. If it cannot, are there still advantages to the EEGMamba multi-task approach?\n\n3. The authors should discuss the relatively low classification accuracy on the motor imagery datasets, which is currently too low for practical motor imagery classification. If this is due to the evaluation setting, additional experiments with per-subject models should be conducted to assess performance, and these results should be compared to other models.\n\n4. Statistical tests are needed to confirm whether the observed differences between models or modules are significant. For instance, statistical analysis should be conducted for the results in Figure 1 comparing EEGMamba and Single-task EEGMamba, as well as for the different ablation models in Figure 6." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Could you please specify how the EEG tasks are encoded into task tokens?\n2. I noticed that the DEAP dataset was utilized in this study, but only data from four electrodes were selected. What is the rationale behind this choice? Additionally, regarding the binary classification on the DEAP dataset, does it pertain to valence, arousal, liking, or dominance? Furthermore, in Table 1, the authors provide the optimal segment lengths for all datasets. What references were used to determine these durations? I observed that, unlike most existing works that employ shorter segments of 1s, 2s, or 4s for the DEAP and SEED datasets, this paper utilizes segment lengths of 60s and 20s. What is the reasoning for selecting such unusually long data lengths?\n3. This work employs five-fold cross-validation for data partitioning, which does not appear to be a commonly used EEG dataset partitioning method. What is the rationale or basis for this choice?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "This paper presents a novel multi-task EEG classification model that incorporates the Bidirectional Mamba and MoE modules, offering a structurally innovative approach. Experiments were conducted on eight datasets, covering a wide range of downstream EEG tasks. The paper is clearly written and easy to follow, with a particularly well-explained method section." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "To address the issues of quadratic computational complexity in handling long-term dependencies in EEG classification models, and the lack of cross-task generalization as most models are designed for single tasks, this paper proposes the EEGMamba model. The model introduces the ST-Adaptive module to address the problem of varying EEG signal lengths and channel numbers across different datasets. It also proposes the Bidirectional Mamba to improve computational efficiency and incorporates the MoE (Mixture of Experts) module to simultaneously capture both the commonalities and differences in EEG signals. Relevant experiments were conducted on eight datasets across four EEG tasks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "This paper proposes a multi-task EEG classification model aimed at performing classification across multiple datasets (downstream tasks) using a single model. However, the experimental results do not demonstrate a clear advantage over other single-task models, making it difficult to convincingly argue for the benefits and necessity of the multi-task approach. Additionally, the ablation study results for the various modules lack significant and consistent differences, making it challenging to prove the effectiveness of each module. Moreover, the motivation for using the Bidirectional Mamba is insufficiently justified. The ST-Adaptive method, proposed to address the issue of varying EEG channel numbers across datasets, is essentially a module integration, lacking in innovation.\n1. In this work, the ST-Adaptive module applies a different one-dimensional convolution for each task, transforming the varying original channel numbers into a fixed number of $D$ channels. However, this approach does not seem to fully achieve the concept of being \"adaptive.\" If a new task emerges with a number of channels outside the predefined range of $C_0$ to $C_N$ in the model, how would this be addressed? This raises concerns about the generalizability and flexibility of the current method in handling unforeseen tasks with different channel configurations.\n2. The purpose of using Mamba in this paper is to reduce computational complexity, conserve resources, and improve efficiency. Generally, the ultimate goal of improving efficiency in EEG models is to achieve real-time recognition. Given that EEG signals, like natural language, possess temporal characteristics, theoretically, a unidirectional Mamba would better meet this requirement, as it only requires past data rather than future information. The motivation for employing Mamba, especially Bidirectional Mamba, in this work is not sufficiently clear or logically aligned with this objective.\n3. In Section 4.1, the authors present the performance of the single-task EEGMamba and other transformer-based models concerning memory usage and inference speed as the sequence length increases. However, it is not clearly evident from Figure 4 that EEGMamba demonstrates a significant advantage over the other methods. Additionally, while it is acknowledged that memory usage increases and inference speed decreases with longer sequence lengths for both single-channel and multi-channel scenarios, the authors do not specify the actual sequence lengths employed in the current eight EEG tasks. This omission lacks a reference point, making it difficult to ascertain whether EEGMamba exhibits superior performance. Furthermore, as indicated in Appendix I, EEGNet appears to perform better in terms of memory usage and inference speed, while also demonstrating commendable performance across various datasets. This further undermines the effectiveness of the proposed method in this paper.\n4. In Section 4.2, the authors present the performance of EEGMamba in multi-task classification. However, I observe that EEGMamba does not demonstrate a significant advantage over the baseline models, and in many datasets, its performance is inferior to that of other single-task models, indicating that the multi-task approach does not facilitate mutual enhancement among tasks. Therefore, I question the necessity of employing a single model to address multiple tasks rather than utilizing several smaller models for different tasks, which might yield better results. The existing findings lack persuasiveness and do not adequately support the motivations for this work or the claims regarding the strong generalization capabilities of the proposed multi-task model.\n5. Regarding Figure 5, I observe that, apart from the sleep stage task, where there is a considerable variation in the activation probabilities of different experts, the activation probabilities for the other tasks across the eight experts are generally quite uniform. This uniformity makes it challenging to demonstrate a particular preference for any specific expert. How can the effectiveness and necessity of the MoE approach be substantiated under these circumstances?\n6. Figure 6 presents the ablation study results for the various modules of EEGMamba. However, the data indicate that these modules appear to have minimal discernible impact, as the experimental results across the Siena, CHB-MIT, and SHHS datasets show little variation. This raises concerns regarding the ability to substantiate the effectiveness of each module." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024eegmamba,\ntitle={{EEGM}amba: Bidirectional State Space Model with Mixture of Experts for {EEG} Multi-task Classification},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=13PclvlVBa},\nnote={under review}\n}" }, "abstract": { "value": "In recent years, with the development of deep learning, electroencephalogram (EEG) classification networks have achieved certain progress. Transformer-based models can perform well in capturing long-term dependencies in EEG signals. However, their quadratic computational complexity poses a substantial computational challenge. Moreover, most EEG classification models are only suitable for single tasks and struggle with generalization across different tasks, particularly when faced with variations in signal length and channel count. In this paper, we introduce EEGMamba, the first universal EEG classification network to truly implement multi-task learning for EEG applications. EEGMamba seamlessly integrates the Spatio-Temporal-Adaptive (ST-Adaptive) module, bidirectional Mamba, and Mixture of Experts (MoE) into a unified framework. The proposed ST-Adaptive module performs unified feature extraction on EEG signals of different lengths and channel counts through spatial-adaptive convolution and incorporates a class token to achieve temporal-adaptability. Moreover, we design a bidirectional Mamba particularly suitable for EEG signals for further feature extraction, balancing high accuracy, fast inference speed, and efficient memory-usage in processing long EEG signals. To enhance the processing of EEG data across multiple tasks, we introduce task-aware MoE with a universal expert, effectively capturing both differences and commonalities among EEG data from different tasks. We evaluate our model on eight publicly available EEG datasets, and the experimental results demonstrate its superior performance in four types of tasks: seizure detection, emotion recognition, sleep stage classification, and motor imagery. The code is set to be released soon." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "EEG Classification", "State Space Models", "Mixture of Experts", "Brain-Computer Interfaces" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/da4a2c7a865952271bb3f9234697bb1f0679448a.pdf" }, "presentation": null, "primary_area": { "value": "applications to neuroscience & cognitive science" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "EEGMamba: Bidirectional State Space Model with Mixture of Experts for EEG Multi-task Classification" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. It has been mentioned that \"minimizing the state-action value function in early training ...\". Does the algorithm considers actions with minimum value \"only\" in early training and does the value of $\\epsilon$ in Algorithm 1 gradually reaches to zero? What is $e$ in algorithm 1?\n\n2. Is there any motivating factor to cluster the games in figure 4 or is it just because of the value range?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. In my view, the main strength of this work lies in the presented theoretical assessment. It shows that the minimum-action value leads to higher temporal difference (TD) than random actions and the difference in the TD is equal to the disadvantage gap. Such finding reveals the underlying importance of the bad actions that may help in accelerate the learning which if often ignored. \n\n2. This work nicely formalizes and defines the relevant concepts, and then gradually presents the core propositions. I have found the paper easy to follow. Also, the detailing of the propositions for both single and double Q-learning is helpful for the reader." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This work considers the problem of experience collection through behavior policy in RL. They argue the benefit of leveraging extremum actions to learn optimal policy and outlined an algorithm that collects and uses such samples. They theoretically show that how actions with minimum value could be helpful. Experimental validation of the approach has been conducted using the Atari game environment." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. While the paper presents several experimental results on ALE, it lacks experiments across different benchmarks. It needs rigorous validation to uphold the claim.\n\n2. Comparison with more recent and effective exploration techniques are missing. \n\n2. Some part of the writing needs improvement. For example, it is very hard to identify how figure 2 and 5 differ." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 2 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "I see the following minor issues and make some auggestions:\n\nLine 50: I wouldn't call $\\epsilon$-greedy \"naive and standard technique\"\n\nLine 96: comma at the end, not a full stop\n\nLine 113: I believe you could use different subscript for $\\theta$ to differentiate the gradient step from the environment time step.\n\nLine 123: full stop at the end of eqn\n\nLine 124: \"a family of algorithms have been proposed based on counting state visitations\" what are these algorithms? I would strongly recommend citing \"R-max – A General Polynomial Time Algorithm for Near-Optimal Reinforcement Learning\" and \"Reinforcement Learning in Finite MDPs: PAC Analysis\" here.\n\nLine 155: Why is there an $s' \\sim \\mathcal T(s, \\hat{a})$ in the first expectation? I don't see any dependence on $s'$ in the term inside the bracket. Same question for later versions of smoothness too.\n\nProof of Proposition 3.4: could you expand on the second inequality please?\n\nLine 264: I am unclear on what is the \"information gained\" here? Is it in an information theoretic sense or in terms of optimizaing the loss?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper provides a good set of experimental results in the low data regime. The method requires fairly simple changes to existing algorithms and it tends to improve performance while being so. The paper is largely well written and I was able to follow along easily." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors present a new algorithm which explores by choosing the worst action as estimated by the neural q function. They demonstrate the efficacy of this in low data regimes for double DWN and compare it to vanilla epsilon greedy based double DQN." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "I see a few important issues that need addressing before I can raise my score.\n\nIn **Proposition 3.4 and 3.6** you start with statements for state $s_t$, which is random variable corresponding to state at time $t$ but in the inequality on the RHS you somehow have $\\mathcal D(s)$ for a fixed state $s$. I am unclear as to where this $s$ is coming from. Moreover, I believe this would significantly complicate the proofs because you will have to account for the time step or \"loosen\" the lower bound because you will have to take some kind of infimum.\n\nWhile I am able to follow intuitively why you would benefit from taking the value minimizing action, I believe you should also include a **comment on the estimated regret** for this choice. It might be beneficial for a randomly initialized $Q$-function in a game setting but we must consider cases where large negative rewards are \"harmful\" to the agent. This is one of the reasons why minimizing regret is important to both theoreticians and practitioners.\n\nIn the experimental section I would be interested to see **comparison with two other papers**: \"Exploration with random network distillation\" and \"Flipping Coins to Estimate Pseudocounts for Exploration in Reinforcement Learning\". Former is based on quantifying how novel a data point is and the latter is directly related to optimistically choosing an action based on pseudo counts. You refer to the inability of doing count based exploration (as done in tabular setting) in your paper but these works are doing some form of counts based exploration. For reference, in lines 126-128 you write\n>> incorporating these count-based methods in high-dimensional state representation MDPs requires substantial complexity including training additional deep neural networks to estimate counts or other uncertainty metrics\n\nI would expect some comparison to how much better these more complex methods are." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Section 5 suggests that experiments were also done with Double DQN (with PER), however all I could find were learning curves for QRDQN, and Table 1 with results for DDQN. Generally, figuring out what results are based on which basis algorithm was challenging as I had to go back to the text several times without success. Could you please clarify what basis agent/algorithm each plot/table is based on. E.g. Figure 3's caption says \"MaxMin TD Learning and canonical temporal difference learning in [ALE]...\". The canonical TD learning method actually implies more of the TD($\\lambda$) class of methods for prediction than a deep RL method such as DDQN or QRDQN. So please specify.\n\n2. Why are the curves for Atari 200M truncated in some cases? (Could be beneficial to add the performance curves for the full length of the experiments.)\n\n3. What was the reasoning behind choosing the specific subset of games for the Atari 200M experiments. \n\n4. Can you comment on the counterexample that I've mentioned in the \"Weaknesses\" section? What is your view on it? (Perhaps experimenting with such a setting would be useful.)\n\nMinor suggestions:\n- Line 87: \"MDP [...] contains continuous set of states\"; I believe this intro is incorrect and also not applicable to the setting of this paper. In Atari and Chain MDP, states are in fact discrete. In Atari, pixel values are discrete, yielding a discrete combinatorial set of states.\n\n- Line 89: The definition corresponds to the *expected* reward function. \n\n- Line 90: The PMF-based definition of the policy does not hold fully for the continuous-state definition of the MDP. But this will be fine if Line 87 is changed to discrete set of states. \n\n- Line 102: I believe the second expectation is mis-specified and in fact is not needed. \n\n- Line 108: \"In deep reinforcement learning, the state space or the action space is large enough that it is not possible to learn and\nstore the state-action values in a tabular form.\"; state-action spaces being large is not a property of DRL. I think better phrasing would be in this line: domains often tackled with DRL tend to have large state and/or action spaces.\n\n- Definition 3.3 seems to be formalized such that $\\theta$ is a random variable of the expectation, but the wording seems to imply that $Q_\\theta$ is a given.\n\n- Would be good to have a visualization of the Chain MDP for ease of readability. Also, what was the number of states $N$?\n\n- Number of environment interactions are not equal to the number of frames in Atari 2600 tasks, because of frame skipping of > 1 used. As such, the X axis labels should change to number of frames.\n\n- The proposed approach is only compatible with discrete-action Q-based methods. That is to say, methods like DDPG cannot utilize it. I think it would be good to mention this somewhere." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "**Interesting problem scenario:**\nConsidering exploration strategies that are directly useful for structural/temporal credit assignment is an interesting area to focus on in approximate/deep RL.\n\n**Analysis tools around acting uniformly vs. Q-minimizing actions after parameter initialization:**\nI found the approach of the propositions to analyze the impact of acting uniformly vs. taking the Q-minimizing action on the TD error to be interesting.\n\n**Experimental testbeds:** \nThe choice of testbeds, ranging from a toy MDP problem to 100K and 200M Atari benchmarks is reasonable. \n\n**Evaluation metrics:**\nReporting Median and 80% aggregate measures, using 5 seeds per each method in Atari for DQN-based methods, and reporting standard error of the mean return are all reasonable choices. However, statistical measures introduced by Agrawal et al. (2021) [arXiv:2108.13264] would have been a step up." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper argues for an alternative exploration strategy to $\\epsilon$-greedy for deep/approximate value-based RL methods (mostly those founded on Q-learning) in which instead of sampling uniformly at random with probability $\\epsilon$, their approach samples actions based on $min_a Q(s, a)$. Algorithmically, the TD update rule of the basis algorithm remains intact. The authors argue that experiences generated by this method of acting have meaningful implications during experience replay/consolidation by TD methods (in particular, deep Q-learning family of algorithms). As such, the authors frame their proposal as a TD approach, in what they call MaxMin TD learning.\n\nThey examine the learning performance on a few tasks of the Atari suite (200M frames), full set of the Atari 100K benchmark, and an illustrative Chain MDP task. The key Atari results are based on the combination of their MaxMin TD learning with QRDQN (a distributional RL algorithm) in comparison with QRDQN with $\\epsilon$-greedy, where MaxMin TD variant achieves higher AUC on both the Median and 80th Percentile aggregate measures." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- **Framing the approach as a TD method, as opposed to an exploration strategy:**\nFraming of the approach as a TD algorithm is not justifiable. A strategic exploration approach could facilitate credit assignment, but categorizing them as a TD approach is rarely ever useful in my view. The proposed method only touches experience generation and not experience consolidation and in this way, I see it as best described as a behavior/exploration strategy. Also, the baselines in question are Noisy Nets and $\\epsilon$-greedy, which are both known as exploration/behavior strategies.\n\n- **Propositions and proofs do not deliver** a full picture of what's going on, unlike the claims for theoretical foundations on par with those existing in tabular settings. \n\n- The approach of only choosing actions from $max Q$ and $min Q$ can easily be shown to introduce bias in a simple counterexample. Say in a multiarmed bandit, action *a* is initialized to the minimal value at random (wrt. to the other initialized actions' values) but as it happens it's *true* Q value is lower than the initialized value. Let's assume also that action *b* is initialized to the maximal value at random (wrt. to the other initialized actions' values) and its corresponding *true* action value is higher than all other initialized actions' values. Note that, even if we use functional approximation (e.g. a neural network) to solve this problem, with parameters shared between the Q estimators for the various actions, it can easily end up being the case that no other actions are experienced during the course of training interactions. This would hold even despite the fact that neither of actions *a* and *b* would be the Q-minimizing or the Q-maximizing actions, respectively, wrt. the *true* Q function.\n\n- Consider the example above again: The TD error reaches zero/near-zero for the Q-minimizing and Q-maximizing actions after a few updates (where Q is the current estimator and not the true Q-function). However, all other actions will have a much higher TD error since no updates is performed on them. This effectively shows that while the results of the Propositions in the paper could hold probabilistically assuming uniform initialization of the outputs of the Q-estimator network, they do not hold on a case by case setting nor do would they hold after several updates (after the uniformity of the outputs is no longer the case).\n \n- Results on Atari 100K are significant, but not on Atari 200M experiments (especially given the fact that the plots for the latter are truncated earlier than 200M frames; e.g. StarGunner is truncated at 70M frames). This likely has ties to my argument above. MaxMin TD could help at the beginning of training (assuming settings like my counterexample occur less commonly in practice / in these tasks), but would not be able to reach higher final performances after enough training of a good baseline on the task. \n\n- I think any benefit emerging from MaxMin TD could have ties to epistemic uncertainty minimization. I think discussions, detailed analysis, and comparisons with approaches directly purposed to do so (incl. bootstrapped DQN of Osband et al., 2016) would have been beneficial." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Q1: Could the authors clarify the decay rate for the hyperparameter $\\textit{Exploration epsilon decay frame fraction}$? It seems that $\\epsilon$ decreases every 0.8% of the total interaction budget. How was this value selected?\n\nQ2: Could you provide more details on how NoisyNetworks was implemented in the experiments? Clarifying architecture choices and how this was selected would be useful, as it apparently allows for a range of configurations.\n\nQ3: In Figure 1, constant 2 appears to balance exploration and exploitation in the UCB algorithm. Has this constant been optimized for the problem, and if so, could the results for varying values be shown? I'd like to see results for values such as [0.5, 1, 2, 3] as done for the epsilon value.\n\nQ4: Could the authors clarify the intended interpretation of Figure 4 for the reader? What is exactly the meaning of TD error being stable or suffering from a drop during environment interactions? What are \"high\" negative or positive values in this case?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The authors tested their approach using both DDQN and a more recent approach that represents the possible returns as distribution (QRDQN), which shows the flexibility of applying MaxMin TD Learning over a variety of different algorithms. Moreover, the tested environment is the well-known Atari benchmark, offering tasks with various characteristics. The paper is relatively easy to follow." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose MaxMin TD Learning, an algorithm that alternates between optimistic and pessimistic strategies for action sampling and Q-estimation updates. This approach addresses sample inefficiency in deep reinforcement learning (RL), though the method offers only incremental novelty. The authors provide both theoretical and empirical analysis, evaluating their method on the popular Atari benchmark." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "* *Marginal novelty*: The proposed method introduces limited novelty since exploring different selection criteria based on Q-estimations has been previously explored with ensembles [1, 2]. Additionally, similar work addressing the optimistic/pessimistic policy update trade-off exists using a more task-dependent strategy [3]. A Related Works section would help clarify where the proposed method advances existing literature. Furthermore, count-based exploration strategies should be referenced in the Background section for completeness.\n\n* *Evaluation in high-dimensional MDPs*: The evaluation lacks depth, particularly concerning high-dimensional MDPs. MaxMin TD Learning is designed to enhance sample efficiency via exploration, yet it is compared against a standard $\\epsilon$-greedy strategy, which performs well given a larger interaction budget and appropriately tuned decay factors. Limiting the interaction budget significantly impacts $\\epsilon$ decay and policy performance, and it appears that the decay factor used here converges too rapidly to its minimum (see Q1). I would recommend including experiments with varying $\\epsilon$ values, especially in the 200-million-frame setting. Additionally, while the 100k benchmark used the NoisyNetworks exploration strategy, it was absent in the 200-million-frame experiments.\n\n* *Fair comparison*: A more balanced comparison would be to benchmark MaxMin TD Learning against alternative approaches designed to enhance sample efficiency as seen in [4, 5]. The authors could emphasize the benefit of MaxMin TD Learning, such as enabling effective learning without requiring prior logged data or a guide policy, which could potentially lead to distribution shifts.\n\n**General remarks**\n\n- In the phrase “Thus, in high-dimensional complex MDPs…”, the citation of Kakade (2003) seems out of place, as deep reinforcement learning was developed later.\n\n- The second question raised saying that the goal is to achieve a *zero cost* experience collection seems infeasible in the context of exploration since interactions with the environment have an inherently associated cost. I think the authors suggest *zero additional cost*\n\n- I suggest having a single Reference section.\n\n**References**\n\n[1] Lan, Qingfeng, et al. \"Maxmin q-learning: Controlling the estimation bias of q-learning.\" arXiv preprint arXiv:2002.06487 (2020).\n\n[2] Agarwal, Rishabh, Dale Schuurmans, and Mohammad Norouzi. \"An optimistic perspective on offline reinforcement learning.\" International conference on machine learning. PMLR, 2020.\n\n[3] Moskovitz, Ted, et al. \"Tactical optimism and pessimism for deep reinforcement learning.\" Advances in Neural Information Processing Systems 34 (2021): 12849-12863.\n\n[4] Yao, Yao, et al. \"Sample efficient reinforcement learning via model-ensemble exploration and exploitation.\" 2021 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2021.\n\n[5] Uchendu, Ikechukwu, et al. \"Jump-start reinforcement learning.\" International Conference on Machine Learning. PMLR, 2023." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024counterintuitive,\ntitle={Counterintuitive {RL}: The Hidden Value of Acting Bad},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=14E7S17hFv},\nnote={under review}\n}" }, "abstract": { "value": "Learning to make sequential decisions solely from interacting with an environment without any supervision has been achieved by the initial installation of deep neural networks as function approximators to represent and learn a value function in high-dimensional MDPs. Reinforcement learning policies face exponentially growing state spaces in experience collection in high dimensional MDPs resulting in a dichotomy between computational complexity and policy success. In our paper we focus on the agent’s interaction with the environment in a high-dimensional MDP during the learning phase and we introduce a theoretically-founded novel method based on experiences obtained through extremum actions. Our analysis and method provides a theoretical basis for effective, accelerated and efficient experience collection, and further comes with zero additional computational cost while leading to significant acceleration of training in deep reinforcement learning. We conduct extensive experiments in the Arcade Learning Environment with high-dimensional state representation MDPs. We demonstrate that our technique improves the human normalized median scores of Arcade Learning Environment by 248% in the low-data regime." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Counterintuitive", "reinforcement learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/e75480b8caf35924aac053b8d7439244b4b39661.pdf" }, "presentation": null, "primary_area": { "value": "reinforcement learning" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/d6a230ee09fcc32b21b8e227e1f6b6335f201f4d.zip" }, "title": { "value": "Counterintuitive RL: The Hidden Value of Acting Bad" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "None" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "No additional questions. Please see the \"Weaknesses\" section for areas needing clarification.\n\n### Recommendations for Improvement:\n- **Refine Prompt Design Explanation:** Providing specific strategies or insights on prompt design tailored for VTG tasks would enhance the paper's originality and usefulness for future researchers.\n \n- **Explore Custom Scene Parsing Techniques:** Introducing refined parsing methods could strengthen TRACE's robustness and accuracy in multi-modal alignment.\n\nThis structured feedback should provide the authors with a comprehensive view of the strengths and areas for enhancement in their paper on TRACE." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper proposes TRACE, a framework leveraging causal event modeling to generate structured video representations through large language models (LLMs). This approach addresses structural gaps in video data, making it valuable for multi-modal research and practical applications in video analysis.\n \n2. TRACE maximizes the potential of pre-trained LLMs by adopting causal event modeling, which decomposes video inputs into frames and aligns them with textual prompts. The temporal segmentation and alignment methods allow videos to be broken down into events with associated timestamps, salient scores, and captions. This granularity is crucial for precise video event parsing and presents a significant step forward in video understanding with LLMs.\n\n3.TRACE outperforms the existing Video-LLMs on three pivotal Video Temporal Grounding (VTG) benchmarks—Charades-STA, QV Highlights, and YouCook2—underscoring its efficacy and robustness in handling video temporal grounding tasks. This achievement underscores TRACE's ability to accurately capture and model the intricate temporal dynamics across a spectrum of video datasets." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a new method for Video Temporal Grounding (VTG) tasks, named TRACE. TRACE uses a causal event modeling framework to represent videos as a sequence of events with timestamps, salient scores, and textual descriptions. The paper designs a task-interleaved video large language model to address the limitations of traditional video LLMs in handling the inherent structure of videos. The TRACE model utilizes different encoders and decoding heads to process visual frames, timestamps, and text inputs, enabling more effective event sequencing and causal modeling. Experimental results demonstrate that TRACE achieves state-of-the-art zero-shot performance on various VTG tasks and, after fine-tuning, can match the performance of traditional non-generative, task-specific models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tWhile causal event modeling is presented as a core contribution of this work, the related work section does not address any prior research on similar methodologies. It would be helpful to clarify whether comparable approaches have been explored in the field of video understanding, or if this approach is entirely novel within this domain. Providing this context could strengthen the argument for the method’s originality and situate it more clearly within existing research.\n\n2.\tIt is unclear whether compressing visual features to 8 tokens is sufficient for preserving critical information in complex video scenes. The paper does not provide an analysis or experimental results on the trade-off between the number of tokens and model performance, which would be valuable in understanding the potential impact of this compression choice.\n\n3.\tThere are several grammatical and spelling errors throughout the manuscript, which impact readability and may detract from the paper’s clarity. For example: Line 22: \"processes\" should be corrected to \"process\". Line 44-45: The phrase \"...which,...\" should be rephrased, and \"lacks\" should be changed to \"which lack\"." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The presentation and illustration are quite clear and easy to follow.\n2. The motivation of causal event modeling is quite intuitive and the design is straightforward and yet effective.\n3. The zero-shot performance is superior compared to previous video LLM methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a task-interleaved video LLM, TRACE, which incorporates a newly-designed causal event modeling framework for VTG task. The TRACE employs multiple encoders for different inputs, while the task tokens are arranged in an interleaved manner. TRACE demonstrates SOTA performance on various VTG datasets compared to previous video LLMs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. While the paper compares TRACE with other video LLMs, it presents limited comparison and may not adequately address how it stands against traditional non-generative and task-specific models.\n2. The extent to which TRACE can be applied to other types of video tasks beyond VTG is unclear. Its design may be highly specialized, which could limit its applicability across diverse video understanding tasks. Authors should present more results on other video-understanding tasks since the design seems generalizable by building such causal event relations." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See Weaknesses" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1) Video Temporal Grounding (VTG) is a crucial task, yet current Video-LLMs underperform in this area. Techniques aimed at improving temporal grounding for these models are highly valuable to advance the field.\n2) The causal event modeling framework fits well with the next-token prediction paradigm of large language models (LLMs), offering an intuitive way to model video structures in sequential tasks.\n3) TRACE demonstrates consistent performance improvements over prior Video-LLMs across three key VTG benchmarks (Charades-STA, QVHighlights, and YouCook2), underscoring its effectiveness." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper addresses the task of Video Temporal Grounding (VTG) and introduces TRACE, a task-interleaved Video-LLM designed for enhanced VTG performance. The authors highlight limitations in current Video-LLMs, which rely solely on natural language generation without considering the inherent temporal structure of videos. To address this, they propose a novel causal event modeling framework, decomposing videos into sequences of events defined by timestamps, salient scores, and textual captions. Extensive experiments on datasets such as Charades-STA, QVHighlights, and YouCook2 demonstrate the superior zero-shot performance of TRACE compared to existing Video-LLMs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1) While the motivation for TRACE is clear, the use of multiple task-specific heads may limit the model’s generalization. A primary appeal of Video-LLMs lies in their ability to handle a variety of tasks without specific fine-tuning. TRACE’s focus on VTG may narrow its versatility, making it less effective for general video understanding tasks. In most cases, lightweight VTG-specific models with stronger performance could be more suitable for VTG scenarios.\n2) Some clarity is not clear. For example, the paper does not adequately explain slot-based compression, which is not a widely known technique. Moreover, compressing each frame to just 8 visual tokens might lead to significant information loss, raising concerns about the trade-off between efficiency and accuracy.\n3) It is unclear whether the same set of number tokens is used for both timestamps and scores. If so, this could blend the two types of information, contradicting the authors' claim (lines 45–46) that the model preserves the distinct structure of video events." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "My main concern is with the autoregressive modeling approach, and if the authors can provide a reasonable explanation, I am willing to consider raising my score, as I believe this work could provide inspiration for future work." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The author first employs a causal modeling method in the grounding of VLLM, achieving causal probability modeling through the sequence of input tokens. This approach will provide inspiration for future work on video understanding tasks using VLLM." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a new Video Temporal Grounding (VTG) method that addresses the shortcomings of existing LLM in handling VTG tasks by modeling causal events." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. **Autoregressive modeling.**\n\nOne of my major concerns is that the authors have only used the earlier events $e_{1:k-1}$ in their modeling of causal relationships between events through autoregression, without incorporating the equally known $e_{k+1:K}$. I believe this approach may be unreasonable since it is likely that the same events may occur earlier while the current event is different due to unrelated pretexts. However, this issue can be avoided by modeling different subsequent events simultaneously. Besides, most current video understanding researchers have modeled multiple events by utilizing all contextual events that occur before and after them [1-4]. This may require the authors to provide further explanation.\n\n[1] Liang C, et al. Visual abductive reasoning[C]. CVPR 2022.\n\n[2] Lam T E, et al. CausalChaos! Dataset for Comprehensive Causal Action Question Answering Over Longer Causal Chains Grounded in Dynamic Visual Scenes. NeurIPS 2024.\n\n[3] Chen T, et al. MECD: Unlocking Multi-Event Causal Discovery in Video Reasoning. NeurIPS 2024.\n\n[4] Du Y, et al. Towards Event-oriented Long Video Understanding. ArXiv 2024.\n\n2. **Inference speed.**\n\nThe authors have adopted a form similar to autoregression, and I would like to understand if there is a time overhead in comparing their model's inference speed to that of current mainstream LLMs.\n\n3. **LLM backbone.**\n\nI noticed that the authors used Mistral-7B as the LLM backbone, however, in other comparison methods, Timechat used LLaMA-2, while HawkEye, Momentor, and VTimeLLM used Vicuna. I would like to know if the authors have conducted experiments with LLaMA-2 or Vicuna as the LLM backbone, to ensure that the superior performance is not due to the better LLM backbone but rather the causal modeling." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "This paper aims to address the mismatch between video structure and video LLMs on video temporal grounding tasks, and propose a causal event modeling framework and the TRACE model as a solution." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024trace,\ntitle={{TRACE}: Temporal Grounding Video {LLM} via Causal Event Modeling},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=14fFV0chUS},\nnote={under review}\n}" }, "abstract": { "value": "Video Temporal Grounding (VTG) is a crucial capability for video understanding models and plays a vital role in downstream tasks such as video browsing and editing. \nTo effectively handle various tasks simultaneously and enable zero-shot prediction, there is a growing trend in employing video LLMs for VTG tasks. However, current video LLM-based methods rely exclusively on natural language generation, lacking the ability to model the clear structure inherent in videos, which restricts their effectiveness in tackling VTG tasks. To address this issue, this paper first formally introduces causal event modeling framework, which represents videos as sequences of events, and predict the current event using previous events, video inputs, and textural instructions. Each event consists of three components: timestamps, salient scores, and textual captions. We then propose a novel task-interleaved video LLM called TRACE to effectively implement the causal event modeling framework in practice. \nThe TRACE processes visual frames, timestamps, salient scores, and text as distinct tasks, employing various encoders and decoding heads for each. Task tokens are arranged in an interleaved sequence according to the causal event modeling framework's formulation.\nExtensive experiments on various VTG tasks and datasets demonstrate the superior performance of TRACE compared to state-of-the-art video LLMs. Our model and code will be made publicly available." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "video large language model", "video temporal grounding" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/e7113f54c3401de1964e3a6b909dacc1224df0f7.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "TRACE: Temporal Grounding Video LLM via Causal Event Modeling" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weaknesses." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The paper has a nice motivation: how does one adapt existing foundation models in order to add an action conditioning to them, so as to make it more relevant and useful for embodied robotics applications\n- The paper writing is clear; first the limitations of prior work are built up and then a solution is proposed" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes a mechanism for adapting current image-conditioned video diffusion models to action-conditioned video diffusion models. They do this by training an additional UNet after the standard video diffusion UNet which predicts an adjustment to the noise output by the standard UNet. Because of this setup, their \"adapter\" does not need access to the parameters of the pretrained video diffusion model. Experiments show that this kind of noise adaptation helps for some metrics and does not for some others." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The way the paper starts with the motivation near L51-52 is a bit misleading. The paper actually cannot fix the issues in L51-52 because they still assume access to the internal inference pipeline of these closed-source model, because if I understand it correctly, this method needs access to a diffusion model's noise prediction at each of the N reverse diffusion steps that happens at inference. For closed source models, this information is not available.\n- The performance gain in the quantitative metrics is not substantial. The metrics where the proposed method shines are mostly photometric quantities. It is not clear if the error margin between prior work and this work just results from the standard deviation or variance of the models. I think a better reflection of the proposed approach would have come from an application to a downstream robot task (maybe manipulation) that would evaluate a robot in action. PSNR and other photometric errors with the shown gain do not say much about the performance of the method.\n- The paper heavily leans on the limitations of the POE approach and that is how a learnable adapter is motivated but qualitatively there is no comparison to that approach (even though POE is slightly better than the action conditioned diffusion across some metrics and settings)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Regarding W1, can authors elaborate more on how and why this analysis motivates design choices in methodology of AVID?\n2. In the ablation study of “No mask” (Table 3), is the adapter trained with $\\epsilon_{\\text{final}}$ given in Equation 5 or the adapter not being able to output a mask?\n3. Since the mask is an important component in design choices of AVID, could author visualize what the mask looks like in different timesteps of diffusion denoising process, which corresponds to Figure 4d?\n4. Since AVID performs two diffusion denoising process, does this increase inference time and thus limit the scope of downstream applications of synthetic videos generated from this approach?\n5. Regarding W3, is it technically possible to apply this approach to domains other than world modeling?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper is well written and easy to follow\n2. The main idea of training a lightweight adapter for action-labeled domains is reasonable. It balances finetuning efficiency and task performance.\n3. Baseline comparisons are comprehensive. Authors compared to many alternative baselines to demonstrate effectiveness of their approach. Authors provide qualitative visualizations for quality of generated videos and usefulness of learned masks." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper focuses on the problem setting of action-conditioned video generation. It proposes to adapt pre-trained video generation model to action labeled dataset without access to parameters from pre-trained models. The goal is to add action information as conditioning to pre-trained models for more accurate video predictions. Authors also analyze limitations in previous related work, production of expert, under a specific case. The proposed approach, AVID, trains an adapter in smaller size on action labeled video datasets. It takes noise predictions outputs from pretrained models and action information as input, and learns to output a mask that is used to combine outputs from pre-trained model and adapter. The adapter is trained with reconstruction loss between final output from both models and ground truth on domain-specific datasets. Authors conducted experiments on two robotics datasets, Procgen and RT1, and compared proposed approach to several baselines that have full access and do not assume access to pretrained model parameters. Experiments results demonstrate that AVID outperforms baselines in generating more realistic videos and better quality given action information on these domains." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The presentation in Section 3.2 is a little unclear. It is hard to connect analysis about limitations of previous work [1] to motivations of the proposed approach\n2. The novelty is somewhat limited. The main difference from previous work is to have domain-specific adapter output an element-wise mask that is used to combine noise predictions from pre-trained model and adapter.\n3. The experimental domains are only two datasets within action-conditioned world modeling\n\n[1] Yang, Mengjiao, et al. \"Probabilistic adaptation of text-to-video models.\" arXiv preprint arXiv:2306.01872 (2023)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "NA" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "see weaknesses." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The authors propose a novel method to condition pre-trained video diffusion models on action sequences without access to the pre-trained model's parameters.\n2. The authors mathematically highlight the limitations of the adaptation method proposed in \"Probabilistic Adaptation of Text-to-Video Models\" and this other approach.\n3. The authors demonstrate that their adaptation method has better action consistency compared to the other approach, using a new metric that they introduce. \n4. The authors also propose multiple baselines to compare against their proposed method." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "1. The authors propose a novel method to condition pre-trained video diffusion models on action sequences without access to the pre-trained model's parameters.\n2. The authors demonstrate that their adaptation method is superior to the method proposed in \"Probabilistic Adaptation of Text-to-Video Models\" and mathematically highlight the limitations of this other approach.\n3. The authors use different pre-trained base models and two different video domains, games and robotics to quantitatively evaluate their proposed method against the above adaptation approach and some other proposed baselines." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. In Table 2, Action conditioned diffusion has a better Action Error Ratio compared to the proposed approach for all three (small, medium, large) variants. While the authors do note this as a limitation, this needs to be explained/investigated more. If it is better to just train an action conditioned diffusion model from scratch why should there be a need to adapt pre-trained models ?\n\n2. Instead of using the action embedding to just scale and shift the t-th frame feature, have the authors explored using cross-attention layers directly with the action embedding sequence similar to language conditioning ? Are there any specific challenges that prohibit such an approach ?\n\n3. It would be interesting to see results for each task type in RT-1 . Are there tasks that are much harder to model than others and what does that tell us about the approach ?\n\n4. Some video visualisations of the generated videos (especially for robotics) would also be very useful to judge the effectiveness of the approach. Are the videos temporally consistent visually ?\n\n5. Why is IRAsim's Action error ratio empty in Table 7 ? is it not possible to evaluate the Action Error Ratio of IRAsim ?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "Most questions and suggestions are detailed in the 'Weaknesses' section.\n\n**Limitations of Naive Adaptation of Yang et al.**\n\nCan the authors please highlight the exact source of discrepancy between the derivation in Yang et al. to the derivation presented in this section? Do you claim that there is an error in their derivation? Alternatively, are there different assumptions in your setting where their derivation does not hold?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "Summarized points:\n- Tackle an interesting problem in the path for scaling robot learning\n- Clear and well written paper\n- Good positioning in related work\n- Comparison to relevant baselines\n- Thorough analysis of results\n- Detailed appendix" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This work proposes a method for leveraging diffusion models pretrained on large-scale action-free video data for training action-conditioned diffusion world models on smaller action-labeled data from domains of interest. The motivation for training these world models is to solve downstream sequential decision-making tasks. The proposed method’s main novelty is that it requires access to some intermediate calculations of the pretrained diffusion model but not to its parameters. The proposed method, AVID, trains an adapter network which is conditioned on the pretrained model’s noise prediction and optimizes a denoising loss that incorporates both noise predictions from the pretrained model and the adapter’s output using a learned mask. The author’s evaluate world model performance on a real-robot and a video game domain based on multiple perceptual metrics as well as an action-prediction-based metric. Baselines include various diffusion-based methods some of which require full access to model parameters. The proposed method either outperforms or is comparable to baselines in most of the evaluated metrics while not requiring access to pretrained model parameters." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Summarized points:\n- Intermediate model calculations are not necessarily more likely to be accessible than the model parameters\n- Limitation analysis of previous work (Section 3.2) does not clearly motivate the author’s specific choice of solution\n- Main motivation is sequential decision-making but evaluation metrics do not assess the world models’ efficacy in solving such tasks\n- It is not clear from the experimental results that training from scratch is not preferable to the proposed method for downstream sequential decision-making\n\n**Evaluation - Metrics**\n\nThe main motivation of your method is to accommodate sequential decision-making but evaluation metrics do not assess the world models’ efficacy in policy learning or planning.\nAll metrics excluding ‘Action Error Ratio’ are perceptual metrics that may be dominated by aspects of the videos that are not important for control. For this reason, I believe the most interesting and relevant metric out of the ones you display in your evaluation is the ‘Action Error Ratio’. Your evaluation could benefit from including additional metrics that are a better proxy for the world model’s usefulness in sequential decision-making. In the Procgen dataset for example, you may want to measure the ability to predict the reward from the generated frames as well as the actions.\n\nI understand that evaluating the world models by actually using them to solve a sequential decision-making task may not be straightforward. Doing this for the RT1 dataset would be hard for multiple reasons, but it may be more feasible for the Procgen environments. One possible evaluation pipeline is training a separate model to predict the reward from a given frame and then use the cross-entropy method (CEM) or a similar sampling-based planning algorithm with model predictive control (MPC) on top of the world model to maximize the sum of rewards in the prediction horizon. Any decision-making algorithm you choose doesn’t have to be SOTA to demonstrate the point that a given world model is better than the other for this purpose.\n\nWhat is the accuracy of the action predictor on each dataset? I believe this is important in order to validate the use of the ‘Action Error Ratio’ metric and that this information should at least be in the appendix.\n\n**Evaluation - Baselines**\n\nWhy do you tune baseline hyperparameters based on FVD and not based on e.g. normalized evaluation metrics? I find this choice puzzling since you explicitly write in the results section that this metric is less suitable than others in the setting of action-conditioned video generation.\n\nHow do you choose which baselines out of the 8 you suggested appear in the result tables?\n\nCan the authors please explain what is the purpose of the ‘Full’ row in the result tables?\n\n**Evaluation - Results**\n\nIt is not clear from the experimental results that training from scratch is not preferable to the proposed method for downstream sequential decision-making, a point that is also suggested in the limitations section and is mostly based on the ‘Action Error Ratio’ metric. This is not to say that it clearly is not beneficial. I suggest adding a discussion about the differences in performance in the two domains which would incorporate further insights as to when and why training an adapter is preferable to training from scratch.\n\n**Evaluation - Ablation Study**\n\n*Mask ablation*: It is not clear from your results that the learned mask has performance benefits and can’t be ‘absorbed’ into the adapter noise prediction, especially since it hurts performance on one dataset and doesn’t in the other.\nHow do you explain the difference in the effects of the mask on performance in each dataset? I think a discussion with respect to factors like the relationship between pre-training and fine-tuning data in each dataset and with respect to the results presented in Figure 4d could shed more light on this matter.\n\n*Conditioning ablation*: I think the method and/or ablation section can benefit from an explanation or intuition behind why conditioning on the pretrained model’s output is beneficial, given that the pretrained output is already accounted for in the objective.\n\n*Request for ablation*: As I see it, the fundamental difference between the proposed method and the PoE baseline is that the parameters of the adapter network are trained on the denoising loss containing noise predictions from both the pretrained network and the adapter network. Therefore an interesting ablation would be combining both the NM and NC ablations." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "Adapting pretrained video diffusion models to action-conditioned world models without finetuning." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024avid,\ntitle={{AVID}: Adapting Video Diffusion Models to World Models},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=15ASUbzg0N},\nnote={under review}\n}" }, "abstract": { "value": "Large-scale generative models have achieved remarkable success in a number of domains. However, for sequential decision-making problems, such as robotics, action-labelled data is often scarce and therefore scaling-up foundation models for decision-making remains a challenge. A potential solution lies in leveraging widely-available unlabelled videos to train world models that simulate the consequences of actions. If the world model is accurate, it can be used to optimize decision-making in downstream tasks. Image-to-video diffusion models are already capable of generating highly realistic synthetic videos. However, these models are not action-conditioned, and the most powerful models are closed source which means they cannot be finetuned. In this work, we propose to adapt pretrained video diffusion models to action-conditioned world models, without access to the parameters of the pretrained model. Our approach, AVID, trains an adapter on a small domain-specific dataset of action-labelled videos. AVID uses a learnt mask to modify the intermediate outputs of the pretrained model and generate accurate action-conditioned videos. We evaluate AVID on video game and real-world robotics data, and show that it outperforms existing baselines for diffusion model adaptation. Our results demonstrate that if utilized correctly, pretrained video models have the potential to be powerful tools for embodied AI." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "world models", "video diffusion", "black box adaptation", "controllable video generation" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/a6cdded6b6e19d7584670b99a59986474e06f242.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/25ac6509dd36691b1b407555902a403283ba7d12.zip" }, "title": { "value": "AVID: Adapting Video Diffusion Models to World Models" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "N/A" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "1. For Figure 3, I wonder if pivot LLM is the original OpenChat-3.5-7B or OpenChat-3.5-7B after fusing training. I also wonder if Target LLM is the OpenChat-3.5-7B after fusing training or the final FuseChat model. Please clarify these.\n2. I wonder if you could categorize the samples in your training data into domains by MT-Bench and see how the distribution is in your training set." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper studies an interesting question of how to fuse multiple chat LLMs into a potent chat LLM. The paper is well-written and well-organized. \n2. The paper has extensive experiments to investigate the effectiveness of their proposed framework and each component in their framework.\n3. Their fusion method is also computation-friendly, which doesn't require additional training or dataset." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a new framework, FuseChat, to fuse diverse LLMs into a single LLM capable of performing various tasks. They first apply pairwise knowledge fusion on source chat LLMs to create multiple target LLMs with identical structures. To fuse models with different vocabulary, they introduce a statistics-based token alignment approach to obtain probabilistic distribution matrices. Then, they fuse all the target LLMs within the parameter space by utilizing the proposed new merging method, SCE. In their experiments, they conducted extensive experiments to investigate their framework with diverse source LLMs and evaluation datasets. They also offered a number of model analyses and ablation studies." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. They didn't provide a significance test to show if their proposed method significantly outperforms their baselines (e.g. FuseLLM/OpenChat-3.5-7B Multi) or not. Because the improvement in some tasks is small, it would be better to show whether the improvement is significant. \n2. Table 1's caption needs to be improved. It would be helpful if they clarified what bold font and underscore mean in their table and what the percentage means." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1.\tHave the authors considered individual model distillation instead of pairwise fusion, given the MinCE choice in Equation 2? \n2.\tWhat is the rationale behind the 0.9/0.1 weight distribution in Equation 4? \n3.\tCan the authors provide statistical significance tests for the improvements over Pairwise Fusion?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1.\tThe motivation is practical and significant, offering a cost-effective solution for integrating capabilities of different heterogeneous LLMs without training new models from scratch.\n\n2. The two-stage framework effectively combines heterogeneous model knowledge through distillation into homogeneous models followed by parameter merging, with a well-designed token alignment strategy.\n\n3. Comprehensive experiments validate the framework's effectiveness, showing competitive performance against different methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces FUSECHAT, a framework designed for the knowledge fusion of chat-based large language models (LLMs). The proposed fuse-and-merge framework integrates the strengths of diverse LLMs through lightweight continual training while avoiding the high cost and potential redundancy associated with developing new LLMs from scratch. Experimental results indicate that the proposed model outperforms existing methods across AlpacaEval and MT-Bench." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper's technical contribution appears somewhat limited. The approach can be viewed as a combination of pairwise FuseLLM and model merging (similar to TIES-Merging), both of which have been previously established as effective methods. The improved performance, while notable, follows logically from the combination of these known techniques, making the technical innovation less impressive than desired.\n2. Several claims in the paper require further clarification. For instance, the statement on line 92 of the Introduction that \"FUSELLM limits its exploration to source LLMs of the same size as the target LLM\" appears inconsistent with FUSELLM's design, which can handle different-sized source models. Furthermore, FUSECHAT doesn't present special designs for distilling from differently-sized source models. Additionally, the choice of MinCE for the Fusion function in Equation 2 reduces to single-model distillation of the model with lower CE score in each pair, raising questions about the necessity of the pairwise approach.\n3. There are concerns regarding experimental details. The combination weight is 0.9 in Equation 4, which means only 0.1 weight is assigned to distillation loss. Compared to 0.9 for SFT, this setting potentially undermines the significance of the distillation process. Moreover, the modest performance difference between FUSECHAT and Pairwise Fusion shown in Table 1 warrants statistical significance testing to validate the improvements." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. First, the challenges of knowledge fusion tasks and the contributions of this paper should be introduced in the Introduction section.\n2. The Method section should highlight the work done by the author. Extensive introduction of work that is not their own will make the article appear less innovative, and you can add formulas to further explain Token Alignment. \n3. The introduction of the SCE algorithm is too short, and the reasons for the use of some steps are not introduced, such as the Calculate and Erase steps.\n4. Added explanations for poor experimental results in the experimental section, for example, Target LLM performs worse than Pivot LLM and Source LLM in some dimensions." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "In general, the logic of the article is good, and the abstract, main text, and conclusions are consistent. The experiments are sufficiently convincing. The author summarizes the previous work from multiple aspects in the related work section." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduce a fuse-and-merge framework called FUSECHAT, which includes two stages. Pairwise knowledge fusion using a pivot LLM and token alignment to generate target LLMs with identical structure and size, and merging these models via SCE method, which determines merging coefficients based on parameter updates." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. In the Introduction section, there is insufficient explanation of the challenges faced by FUSECHAT. It is not enough to just explain the advantages of knowledge fusion, but the complexity of the work should also be highlighted.\n2. The contribution of the work done in this paper is not explained in the Introduction section. \n3. The method section uses too many narrative words and lacks specific formula expressions, which increases the difficulty for readers to understand the article. \n4. In the experiment section, there is a lack of explanation for the adverse results in the experiment." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "In this work, we propose a fuse-and-merge framework for knowledge fusion of structurally and scale-varied chat LLMs to integrate their collective knowledge and individual strengths into a more potent chat LLM, resulting in FuseChat." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024fusechat,\ntitle={FuseChat: Knowledge Fusion of Chat Models},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=15UetYngA7},\nnote={under review}\n}" }, "abstract": { "value": "While training large language models (LLMs) from scratch can indeed lead to models with distinct capabilities and strengths, it incurs substantial costs and may lead to redundancy in competencies. Knowledge fusion aims to integrate existing LLMs of diverse architectures and capabilities into a more potent LLM through lightweight continual training, thereby reducing the need for costly LLM development. In this work, we propose a new framework for the knowledge fusion of chat LLMs through two main stages, resulting in FuseChat. Firstly, we conduct pairwise knowledge fusion on source chat LLMs of varying structures and scales to create multiple target LLMs with identical structure and size via lightweight fine-tuning. During this process, a statistics-based token alignment approach is introduced as the cornerstone for fusing LLMs with different structures. Secondly, we merge these target LLMs within the parameter space, where we propose a novel method for determining the merging coefficients based on the magnitude of parameter updates before and after fine-tuning. We implement and validate FuseChat using six prominent chat LLMs with diverse architectures and scales, including OpenChat-3.5-7B, Starling-LM-7B-alpha, NH2-SOLAR-10.7B, InternLM2-Chat-20B, Mixtral-8x7B-Instruct, and Qwen-1.5-Chat-72B. Experimental results on two instruction-following benchmarks, AlpacaEval 2.0 and MT-Bench, demonstrate the superiority of FuseChat-7B over baselines of various sizes. Our model is even comparable to the larger Mixtral-8x7B-Instruct and approaches GPT-3.5-Turbo-1106 on MT-Bench." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Model Fusion", "Large Language Models" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/129f5d0cbd1af8519a0f0e8e5f6e009d00f5c7fd.pdf" }, "presentation": null, "primary_area": { "value": "foundation or frontier models, including LLMs" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/70909a90c768d4664f54918eff8a8c7125d2ffba.zip" }, "title": { "value": "FuseChat: Knowledge Fusion of Chat Models" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Why is the case of $\\epsilon > 1$ considered interesting for LDP studies?\n\n2. In Proposition 1 (2), to ensure user-level $\\epsilon$-LDP from item-level $\\epsilon$-LDP, if we randomly pick a sample from each user, why is it stated as ''$n$ users with $m$ samples per user'' instead of ''$n$ users with $1$ sample per user''?\n\n3. For Definition 1, could you explain in detail why the definition of user-level $\\epsilon$-LDP does not ensure item-level $\\epsilon$-LDP?\n\n4. For Theorem~1, I am unable to understand why is it said that the mean squared error will never converge to zero with increasing $m$ if $n$ is fixed.\n\n5. What does $n_0$ represent in Equation (6)?\n\n6. For the stochastic optimization problem, why is only the bounded gradient case considered? Why can't the private mean estimation over unbounded support developed in the paper be used for the unbounded gradient case, which seems more interesting and important in practice?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper tackles significant learning problems under user-level local differential privacy (LDP) constraints and establishes several tight lower and upper bounds." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper first analyzes the mean estimation problem and then extends the findings to stochastic optimization, classification, and regression. Specifically, the authors propose adaptive strategies to achieve optimal performance across all privacy levels. They also derive information-theoretic lower bounds, demonstrating that the proposed methods are minimax optimal up to logarithmic factors. Notably, unlike the central DP model, where user-level DP generally leads to slower convergence, the results show that, under the local DP model, convergence rates are nearly identical between user-level and item-level cases for distributions with bounded support. For heavy-tailed distributions, the user-level rate is even faster than the item-level rate." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Some statements throughout the paper are somewhat unclear, which can make parts of the presentation difficult to follow.\n\nFor the stochastic optimization problem, only the bounded gradient case and strongly convex objective functions are considered, which may not be sufficiently practical for broader applications." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Please see above." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper is very organized and presents its results in a clear manner.\n\n2. Matching information-theoretic lower bounds are also derived which enhances the completeness of this work." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper examines user-level privacy in a distributed setting, particularly in user-level local differential privacy (ULDP). The authors analyze mean estimation and its applications in stochastic optimization, classification, and regression, proposing adaptive strategies that optimize performance across various privacy levels. The authors claim that unlike in the central model, the convergence rates for user-level and item-level privacy are nearly equivalent in local models, with user-level privacy yielding even faster rates for heavy-tailed distributions." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "This paper studies ULDP on various problem settings: mean estimation, stochastic optimization, classification and regression. It is clear from Table 1 how the proposed rates in ULDP is different from the rates in item-level LDP. However, some relevant papers appear to be missing from the references. For example, [1], [2] and [3] \n\n[1]: Li, Bo, Wei Wang, and Peng Ye. \"Improved Bounds for Pure Private Agnostic Learning: Item-Level and User-Level Privacy.\" arXiv preprint arXiv:2407.20640 (2024).\n\n[2]: Cummings, Rachel, et al. \"Mean estimation with user-level privacy under data heterogeneity.\" Advances in Neural Information Processing Systems 35 (2022): 29139-29151.\n\n[3]: Charles, Zachary, et al. \"Fine-tuning large language models with user-level differential privacy.\" arXiv preprint arXiv:2407.07737 (2024).\n\n\nBesides, on line 132 and 133\" Moreover, we also provide the first analysis on nonparametric classification and regression problems under user-level ϵ-LDP\" is not accurate. To the best of my knowledge, [4] also studies regression in the ULDP setting under sparsity constraint. From my perspective, sparse estimation problem in LDP model ([5], [6]) might also could also be a valuable addition to the related work section.\n\n[4]: Ma, Yuheng, Ke Jia, and Hanfang Yang. \"Better Locally Private Sparse Estimation Given Multiple Samples Per User.\" arXiv preprint arXiv:2408.04313 (2024).\n\n[5]: Zhu, Liyang, et al. \"Improved Analysis of Sparse Linear Regression in Local Differential Privacy Model.\" arXiv preprint arXiv:2310.07367 (2023). \n\n[6]: Zhou, Mingxun, et al. \"Locally differentially private sparse vector aggregation.\" 2022 IEEE Symposium on Security and Privacy (SP). IEEE, 2022." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. The authors highlight the regime where \\eps > 1 in the introduction. Yet, it is unclear how this regime is handled in the proof of the lower bound. For example, in the proof of Theorem 2, how do we get from Eq. 62 to Eq. 64, if \\eps > 1? I understand that the current proof holds for \\eps < 1. Similar questions exist in the proof of Theorems 6 and 7. \n\n2. Following the above, it would be useful to highlight the difference in the lower bound proof for item-level and user-level LDP, especially the regime when \\eps >1. \n\n3. It seems to me that the current local model is **non-interactive**? Can the authors comment on the **interactive** model? That is, can the proposed algorithms be easily extended to the interactive model?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "+ It addressed user-level DP, which is a relatively less explored but extremely relevant area\n+ It studied a wide variety of tasks (mean estimation, stochastic optimization, nonparametric classification and regression)" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper addresses the problem of achieving user-level local differential privacy (LDP) across various statistical tasks, including mean estimation, stochastic optimization, classification, and regression. By tailoring privacy mechanisms to different privacy levels, the authors propose algorithms that attain optimal performance rates under user-level LDP, achieving minimax optimality up to logarithmic factors. Unlike the central model, where user-level privacy often implies slower convergence, the local model yields convergence rates comparable to item-level LDP, with even faster rates in heavy-tailed distributions. This work provides both theoretical bounds and adaptive strategies, expanding the scope of user-level LDP applications in distributed systems." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Technical novelty is unclear\n- Some proof is unclear" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "A systematic study on user-level local diferential privacy on various tasks including mean estimation, stochastic optimization, classification and regression" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024learning,\ntitle={Learning with User-Level Local Differential Privacy},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=15dVqf7VXR},\nnote={under review}\n}" }, "abstract": { "value": "User-level privacy is important in distributed systems. Previous research primarily focuses on the central model, while the local models have received much less attention. Under the central model, user-level DP is strictly stronger than the item-level one. However, under the local model, the relationship between user-level and item-level LDP becomes more complex, thus the analysis is crucially different. In this paper, we first analyze the mean estimation problem and then apply it to stochastic optimization, classification, and regression. In particular, we propose adaptive strategies to achieve optimal performance at all privacy levels. Moreover, we also obtain information-theoretic lower bounds, which show that the proposed methods are minimax optimal up to logarithmic factors. Unlike the central DP model, where user-level DP always leads to slower convergence, our result shows that under the local model, the convergence rates are nearly the same between user-level and item-level cases for distributions with bounded support. For heavy-tailed distributions, the user-level rate is even faster than the item-level one." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Local differential privacy", "minimax" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/a9ae0d50e13ee4fc6188b1059291ff9ce7b533be.pdf" }, "presentation": null, "primary_area": { "value": "learning theory" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Learning with User-Level Local Differential Privacy" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "N/A" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The idea of converting camera poses into pixel-wise embeddings is novel, which allow the video generation model to effectively understand the camera motion.\n\n2. This paper studies three common ways of incorporating camera pose embedding into a DiT model, which could be useful for future work." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents an approach to enable controlling camera pose for video generation based on Diffusion Transformer (DiT). It converts pixel-wise motion field based on Plucker coordinates into a sparse motion field, which is then injected into the temporal attention part of DiT. LoRA is used to fine-tune a pre-trained DiT model (Open-Sora). Experimental results on the RealEstate10K dataset are reported." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The proposed sparse motion encoding module seems almost identical to the standard pixel-wise Plucker embedding. Compared with Eq. (3), the only difference in Eq. (4) is the embedding computation are performed on a set of sparse locations controlled by $s_x$ and $s_y$. It is not clear how the camera \"motion\" is encoded. Does the proposed approach convert the pixel-wise motion vectors shown in Fig. 4 into embeddings? \n\n2. A lot of definitions are not clear and math symbols are not used rigorously in the presentation, making the paper hard to follow and understand.\n\n- a) The Sparse Motion Encoding Module and Temporal Attention Injection Module are not shown in Fig. 1 at all.\n\n- b) In line #179 on page 4, how is $RT$ defined? Is it matrix multiplication similar to $RK^{-1}$?\n\n- c) In line #195 on page 4, it says $F_s\\in \\mathbb{R}^{L\\times M\\times N}$? According to the definition in Eq.(4), the channel dimension shouldn't be 1?\n\n- d) In Fig. 2, $c_p$, $c_s$, and $c_l$ are not defined in the main text. And the shape of $c_l$ is not clearly explained either.\n\n- e) In Fig. 3, what are $s$, $p$, and $p_m$?\n\n3. The first two items of the claim contributions of the paper are essentially identical. Both of them are about incorporating camera poses into a DiT model.\n\n4. Only visual results of simple camera motion (zoom in, zoom out, and roundabout) are shown in the paper. No supplementary results are available. It is therefore hard to gauge the effectiveness of the proposed approach." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "- Could the authors provide a user study to assess the quality?\n\n- Could the authors upload the generated videos and visual comparisons with baselines?\n\n- Could the authors provide the implementation details of cross-attention and adaptive normalization? Including where these computations happen in relation to temporal attention computation. (In Figure 3, the injection happens between two temporal attention layers. This figure is wrong to me as exactly one temporal attention should exist either before or after the injection.)\n\n- Could the authors provide the reasons why the models are trained on only 16-frame videos?\n\n- Could the authors detail how the model is extended to 72 frames given the model is trained on 16 frames?\n\n- The authors mention that \"object movement tends to be limited to small-scale motions\". I think this can be a big issue. Could the authors provide a detailed comparison with other baselines?\n\n- Is the motivation for introducing sparse sampling of the motion field for computational efficiency? The authors lately argue that sparse sampling improves the results, but I am not convinced why the performance is improved. Could the authors provide more detailed reasons behind that?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- To the best of my knowledge, this is the first work that tackles camera-motion-controlled video generation with open-source DiT (i.e. opensora).\n\n- The idea of sparsely sampling motion fields before inputting them into the VAE encoder is new.\n\n- They demonstrate that adaptive normalization for conditioning camera motions is the effective strategy for camera-conditioned video generation for the first time, which is consistent with the results demonstrated in trajectory-controlled generation (e.g. Tora). \n\n- They quantitatively demonstrate that the generated videos have better motion and visual quality for 72-frame generation." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents an approach to training a DiT-based video diffusion model (OpenSora) to control the camera motions of generated videos. \nThe camera motion, which is represented by Plücker coordinates, is first sparsely sampled (downsampled by x40) and then encoded into the \"motion latent\" via VAE encoder (MegViT-v2, inspired by Tora). Finally, motion latent is injected into the temporal attention layer of DiT via adaptive normalization. The model is finetuned on 16-frame videos from the RealEstate10K dataset. The authors demonstrate that visual quality and motion accuracy (FID, FVD, CamMC) outperformed baselines for 72 frame generation." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Novelty:\n\n- As the authors have already acknowledged, the idea of using Plucker coordinates has already been introduced (e.g., VD3D). Additionally, the use of a VAE encoder and adaptive normalization has already been introduced by Tora. Following that, the main technical contribution is introducing a sparsely sampled motion field. The author argues that sparsely sampled motion fields contribute to performance improvement, but the authors fail to provide details results (e.g., visual results, more ablation study in Table 2, what about x1?) nor detailed motivation. Additionally, choosing this downsample factor seems heuristic with no intuitive justifications. I would appreciate it if the authors could provide more ablation studies and technical motivations for applying sparsely sampled motion fields.\n\nExperiments:\n\n- Although the model is trained on a 16-frame dataset, the model performs worse than other baselines for 16-frame generation. Additionally, the motivation for training only on 16-frame videos is unclear, given the model is tasked to generate longer-frame videos during the experiments. I would appreciate it if the authors could provide more explanations for this decision.\n\n- The authors did not provide sufficient qualitative results or user study, where the superiority of their method is not convincing. \n\nClarity:\n\n- The paper lacks implementation details. For instance, I am not sure how adaptive normalization and cross-attention are performed exactly. Figure 3 seems inaccurate because the injection happens between two temporal attention layers, where the temporal attention layer should exist in only one of these locations. Please see the questions below for more requests for clarification." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Besides the questions in the first point of weakness section, I have the following questions.\n1. The MagVit2 model is designed to be able to compress the images and videos in a single model. Using some padding, the first image is treated as a separate image, thus the training of MagVit2 VAE is 17 frames (line 17) is reasonable. But, in line 259, the author said \"we extract 16-frames...\", I want to know how those 16 frame are padded and what is the output of the VAE encoder?\n2. Can the author provide the reconstruction results, using l1 loss, for the reconstructed sparse Plücker embedding?\n3. Whether the motion degree of objects of the whole scene degraded after adding the camera control-related modules? \n4. In line 303, the author state that they use the different resolution for different video generation models, can the FID, FVD fully reflect the ranking of different models?\n5. In the visualization results, the camera trajectories seems too simple, focusing on panning and zooming. I remember there are some more complex camera trajectories in RealEstate10K dataset, can the author provide some quantitative or qualitative results on those complex camera trajectories?\n6. How to calculate the CamMC metric for SVD, AnimateDiff, OpenSora model, since they cannot take the camera trajectories as input." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper train a VAE model to temporally downsample the Plücker embeddings, easing the conflict of different temporal length between Plücker embedding and the latent features. \n2. The visualization results demonstrate the effectiveness of the proposed method on some simple camera trajectories, like paning and zooming." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper aims at controlling the camera viewpoints of the videos generated by the DiT-based video diffusion models. To achieve the precise camera viewpoint control, this paper utilizes the Plücker embedding as the camera representation. The Plücker embeddings are per-frame spatial maps, while the DiT-based diffusion (like OpenSora) do some downsamples in the temporal dimension. To deal with this conflict, this paper proposes a Sparse Motion Encoding Module to temporally downsample the Plücker embeddings, with the same ratio as the OpenSora VAE. This Sparse Motion Encoding Module is implemented by a MagVit2 like causal VAE. The generated latent motion is injected into the temporal attention layer of OpenSora using an adaptive normalization layer. Experiments demonstrate the superiority of proposed method on both short and long video generation (with camera control) task. Some ablation studies also prove the effectiveness of the proposed Sparse Motion Encoding Module and Temporal Attention Injection Module." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The motivation of Sparse Motion Encoding Module is not well presented. Why the VAE model is used to compress the Plücker embedding? The encoder of VAE can be used to compress the Plücker embedding, what is the decoder used for? Besides, generally, the VAE model will not bring too much extra computation, and it can be reused once encoded, why this module **sparsely** sample some Plücker motion vector?\n2. The writing is not good for this manuscript. For example, there are some typos, like the **MegVit-v2** in Line 198, the inconsistency of OpenSora and Open-Sora. Besides, some details are missing, like what is the input of the Sparse Motion Encoding Module. The rows 2 and 4 in Figure 4 does not provide too much information.\n3. The experiments is not very convincing. See Question part." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "1. In Fig 1, I am curious how to convert the text instruction \"Zoom-in\" to the motion field.\n2. Eq (3) is unclear. Is P_{x,y} the Plucker embedding? But how to get R, K and t from the dataset? Does the dataset provide such information?\n3. What is the major difference between this paper and the motionCtrl [1]? A detailed comparison of the proposed method with MotionCtrl, highlighting key differences in approach, architecture, and performance, would be helpful.\n\n[1] Zhouxia Wang, Ziyang Yuan, Xintao Wang, Yaowei Li, Tianshui Chen, Menghan Xia, Ping Luo, and Ying Shan. 2024. MotionCtrl: A Unified and Flexible Motion Controller for Video Generation. In ACM SIGGRAPH 2024 Conference Papers (SIGGRAPH '24). Association for Computing Machinery, New York, NY, USA, Article 114, 1–11. https://doi.org/10.1145/3641519.3657518" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The authors add the camera pose in the video generation, which is an interesting point." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the authors propose a DiT-based video generation method that embeds the camera poses as controllable signals. They use LoRA to fine-tune the attention layer parameters in the training. RealEstate10K dataset is used in the evaluation." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Many details are missing in Fig 2. I do not find any explanations of what does frame pose mean and how to get the frame pose. How to convert frame pose to camera pose is also very unclear. It would be helpful to provide a step-by-step description of how to extract and process the pose information from the dataset, including definitions of frame pose and camera pose, and the conversion process between them.\n2. The method is only evaluated on a single dataset, which is not sufficient to verify the effectiveness of the method. For example, authors can test on videos from WebVid and HD-VILA following MotionCtrl [1] paper.\n\n[1] Zhouxia Wang, Ziyang Yuan, Xintao Wang, Yaowei Li, Tianshui Chen, Menghan Xia, Ping Luo, and Ying Shan. 2024. MotionCtrl: A Unified and Flexible Motion Controller for Video Generation. In ACM SIGGRAPH 2024 Conference Papers (SIGGRAPH '24). Association for Computing Machinery, New York, NY, USA, Article 114, 1–11. https://doi.org/10.1145/3641519.3657518" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024ccmdit,\ntitle={{CCM}-DiT: Camera-pose Controllable Method for DiT-based Video Generation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=15lk4nBXYb},\nnote={under review}\n}" }, "abstract": { "value": "Despite the significant advancements made by Diffusion Transformer (DiT)-based methods in video generation, there remains a notable gap with camera-pose perspectives. Existing works such as OpenSora do not adhere precisely to anticipated trajectories, thereby limiting the utility in downstream applications such as content creation.\nTherefore, we introduce a novelty approach that achieves fine-grained control by embedding sparse camera-pose information into the temporal self-attention layers. We employ LoRA to minimize the impact on the original attention layer parameters during fine-tuning and enhance the supervision of camera-pose in the loss function.\nAfter fine-tuning the OpenSora’s ST-DiT framework on the RealEstate10K dataset, experiments demonstrate that our method outperforms LDM-based methods for long video generation, while maintaining optimal performance in trajectory consistency and object consistency." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Video Generation", "Diffusion Models" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/1614a2503c1fc1701e82d661c9176a0928e52260.pdf" }, "presentation": null, "primary_area": { "value": "generative models" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "CCM-DiT: Camera-pose Controllable Method for DiT-based Video Generation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 2 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weakness." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The paper is clearly written and organized, with effective figures explaining both W-Bench and VINE.\n\n- The paper provides rigorous evaluations, testing VINE and eleven other watermarking models on diverse editing techniques." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents VINE, a watermarking method designed to withstand various image editing techniques enabled by advanced generative models. It also introduces W-Bench, a benchmark that evaluates watermark robustness against multiple types of edits, making it a valuable resource for watermarking research." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- EditGuard is primarily designed for editing detection, not robust watermarking, and it was not tested with its most robust configuration. This impacts the fairness of the evaluation, as EditGuard’s focus and strengths differ from VINE’s intended use." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Although the watermarking against Image Editing is interesting and novel, I cannot get the value of this task. Can you elaborate the perspective of this task?\n2. The author hypothesizes that a powerful generative prior can facilitate embedding information more invisibly while enhancing robustness (Line 249). Why hypothesize that? What are the assumptions based on?\n3. What is the purpose of finetuning VINE-B to VINE-R using Instruct-Pix2Pix? (Line 323)\n4. Why is the resolution not unified? (Line 1042) \n5. Is VINE only work on the Image Editing task? What about other common watermarking tasks?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1.\tThe proposed method is easy yet effective. The combination of different losses is reasonable.\n2.\tThe validation of watermarking patterns in high-frequency bands after image editing and blurring is solid.\n3.\tThe experimental results show the proposed watermarking method is robust enough against multiple image editing methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces W-Bench, the first comprehensive benchmark designed to evaluate the robustness of watermarking methods against a wide range of image editing techniques, including image regeneration, global editing, local editing, and image-to-video generation. Authors reveal that image editing and blurring distortion predominantly remove watermarking patterns in high-frequency bands, while those in low-frequency bands remain less affected. Based on this, distortions are used as surrogate attacks to overcome the challenges of using T2I models during training and to enhance the robustness of the watermark. The authors approach the watermark encoder as a conditional generative model and introduce two techniques to adapt SDXL-Turbo, a pretrained one-step T2I model, for the watermarking task. Experimental results demonstrate that VINE is robust against multiple image editing methods while maintaining high image quality." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tThis paper lacks the validation of hypotheses in Line 249.\n2.\tThe task of watermarking against Image Editing seems worthless.\n3.\tThe watermarking pattern existing in high-frequency bands after image blurring is not a new discovery. However, the author spends too much text on it." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. I have doubts about the results in Figure 5(a). The experimental results show that 250-step noise in image regeneration can significantly disrupt the watermark（bit acc). Does this mean that global image editing (e.g., SDedit, prompt2prompt) with 250 steps can also completely remove the watermark? If so, I believe this result does not demonstrate robustness, as global image editing often uses even more denoising steps." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Comprehensive Evaluation Framework: W-Bench covers a variety of image editing techniques, providing a comprehensive platform for assessing the robustness of watermarking methods.\n\n2. Innovative Use of Generative Priors: VINE embeds watermarks by adapting pretrained large-scale generative models, making the embedding more imperceptible and robust.\n\n3. This task is innovative, focusing on watermarking that is robust against image editing methods." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces a new evaluation benchmark, W-Bench, designed to test the robustness of image watermarking methods under image editing supported by large-scale generative models. W-Bench includes image regeneration, global editing, local editing, and image-to-video generation. The authors also propose VINE, a watermarking method utilizing generative priors to enhance the robustness and visual quality of watermark embedding. Experiments show that VINE outperforms existing watermarking methods across various image editing techniques." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "TreeRing, Gaussian Shading, and RingID, which add watermarks in the frequency domain of the initial noise, are generally considered robust against image editing (e.g., prompt2prompt) and regeneration. This paper lacks this crucial comparison. If these methods are also robust to image editing, the contribution of this paper may be diminished.\n\nReference:\n1. Tree-ring watermarks: Fingerprints for diffusion images that are invisible and robust\n2. Ringid: Rethinking tree-ring watermarking for enhanced multi-key identification\n3. Gaussian Shading: Provable Performance-Lossless Image Watermarking for Diffusion Models" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. What is the reason for choosing only these four types of image editing methods (image regeneration, global editing, local editing, and image-to-video generation) to evaluate the image watermarking robustness, against image editing? \n2. What is the motivation for using SDXL-Turbo as the generative prior for watermark encoding? If it is just to avoid multi-step sampling, there should be lots of one-step generative models to choose from, for example, the SDXS [2]. \n\n[2] Song, Yuda, Zehao Sun, and Xuanwu Yin. \"SDXS: Real-Time One-Step Latent Diffusion Models with Image Conditions.\" arXiv preprint arXiv:2403.16627 (2024)." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper focuses on the image watermark robustness against image editing, which is important but has rarely been explored.\n2. The proposed benchmark includes different types of image editing approaches, rendering it comprehensive to some extent.\n3. The proposed SDXL-Turbo-based robust image watermarking method is novel, and the experiments demonstrate its effectiveness.\n4. The paper is overall well-written." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces an image watermarking benchmark, specifically aiming to evaluate the watermark robustness against four image editing methods. In addition, an image watermarking that is robust against image editing is proposed." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The benchmark only considers four types of image editing methods (image regeneration, global editing, local editing, and image-to-video generation). Other image editing methods such as style transfer are not considered.\n2. Only one image-to-video generation method is included in the benchmark. The robustness against other image-to-video generation methods such as [1] is not evaluated.\n\n\n[1] Hu, Yaosi, Chong Luo, and Zhenzhong Chen. \"Make it move: controllable image-to-video generation with text descriptions.\" Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1.Figure 6 in the appendix shows that VINE exhibits higher brightness in the central region, providing evidence for why the proposed watermarking method demonstrates strong robustness against image editing. If the author can thoroughly elucidate the principles underlying this phenomenon, it may address the previously mentioned issue of \"a disconnect between the author's analysis of watermark robustness and the design of the watermark model.\"\n\n2.The experimental results demonstrate that the proposed watermarking method, VINE, significantly enhances robustness against various image editing techniques. Has the author considered using representative image editing as an attack template, incorporating the associated attack loss as one of the objective functions during the training phase? Alternatively, how might integrating the specific effects of image editing on watermarks into the design of the watermarking model influence the results of the watermarking algorithm?\n\n3. In the experimental section, some of the differences between the subjective experimental results are difficult to discern visually. The author could consider selecting a subset of images and enlarging specific regions to facilitate reader comprehension." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper presents the first holistic benchmark that incorporates four types of image editing techniques to assess the robustness of watermarking methods. This is significant for evaluating the robustness of future watermarking methods, as it helps to promote the standardization and comprehensiveness of robustness assessments. By addressing a critical gap in evaluating watermark resilience against sophisticated transformations enabled by modern generative models, this work encourages researchers in the field of image watermarking to focus on the robustness of their methods against emerging image editing technologies, including image regeneration, global editing, local editing, and image-to-video generation. Overall, the paper is clearly articulated and well-supported." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper evaluates eleven watermarking methods against prevalent image editing techniques and demonstrates that most methods fail to detect watermarks after such edits. It also introduces a watermarking model based on SDXL-Turbo, which exhibits high robustness against these editing methods while maintaining high image quality." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper explains the reasons behind the watermarking algorithm's resistance to image editing from the perspective of the frequency domain. It notes that the watermarking methods exhibiting high robustness against image editing in certain scenarios display prominent patterns in the low-frequency bands, which aligns with the general understanding of watermark robustness. However, the paper primarily focuses on the robustness of watermarking methods against image editing techniques based on generative models. Therefore, summarizing the unique effects of such image editing techniques on the watermark is more meaningful.\n2. We observe that the proposed watermarking method, VINE, shows higher brightness in the central region of the frequency domain, which corresponds to the author's analysis of watermark robustness. However, the paper does not clarify why this watermarking model based on SDXL-Turbo exhibits such characteristics, leading to the author's specific design of the watermark algorithm. In other words, there seems to be a disconnect between the author's analysis of watermark robustness and the design of the watermark model." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We present the first comprehensive benchmark for evaluating the robustness of eleven watermarking methods against prevalent image editing techniques and propose a watermarking model based on SDXL-Turbo that remains robust to these editing methods." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024robust,\ntitle={Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=16O8GCm8Wn},\nnote={under review}\n}" }, "abstract": { "value": "Current image watermarking methods are vulnerable to advanced image editing techniques enabled by large-scale text-to-image models. These models can distort embedded watermarks during editing, posing significant challenges to copyright protection. In this work, we introduce W-Bench, the first comprehensive benchmark designed to evaluate the robustness of watermarking methods against a wide range of image editing techniques, including image regeneration, global editing, local editing, and image-to-video generation. Through extensive evaluations of eleven representative watermarking methods against prevalent editing techniques, we demonstrate that most methods fail to detect watermarks after such edits. To address this limitation, we propose VINE, a watermarking method that significantly enhances robustness against various image editing techniques while maintaining high image quality. Our approach involves two key innovations: (1) we analyze the frequency characteristics of image editing and identify that blurring distortions exhibit similar frequency properties, which allows us to use them as surrogate attacks during training to bolster watermark robustness; (2) we leverage a large-scale pretrained diffusion model SDXL-Turbo, adapting it for the watermarking task to achieve more imperceptible and robust watermark embedding. Experimental results show that our method achieves outstanding watermarking performance under various image editing techniques, outperforming existing methods in both image quality and robustness. Our model and benchmark will be publicly available." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "AI Security", "Watermark", "Diffusion Model", "Image Editing" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/e9740056026febb22ac684c0b11d0b966af36c8b.pdf" }, "presentation": null, "primary_area": { "value": "generative models" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Robust Watermarking Using Generative Priors Against Image Editing: From Benchmarking to Advances" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "I would be grateful if the authors could respond and address the weaknesses. I am willing to increase my score if the authors could address the weaknesses." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper is relatively easy to follow and well-written.\n2. The proposed \"Twicing Attention\" is simple and easy to implement.\n3. Theoretical motivation and the mathematical details behind their motivation and choices have been provided." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The over-smoothing problem in Transformers is a well-known phenomenon, where the outputs of different attention layers in a Transformer model are highly similar. This paper introduces Twicing Attention to address this problem, which uses low-pass NLM smoothing filters to tackle this problem. The core idea can be phrased as, instead of using the standard attention matrix $A$, to use $2A - A^2$." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper compensates for the simplicity of the core idea by over-explaining and being overly verbose. For example, most of the material on pages 7-8 can be summarised in 2-3 paragraphs. Even Algorithm 1 on page 8 is redundant and too verbose. The algorithm's objective is clear and simple: to compute $2A - A^2$. I don't think one needs 12 lines to explain that.\n2. Instead, the paper could have added to its contribution through a more thorough study. E.g., one avenue for improvement would be to consider other candidates besides the $2A - A^2$ and then compare them in the considered benchmarks" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "My question lies in the efficiency comparison (Tab. 4). Despite the fact that Twicing has the same complexity of $O(N^2 d)$ as claimed in the paper, it still increases the overhead by an additional 50% due to the extra matrix multiplication in line 7, Alg. 1. However, Tab. 4 indicates that implementing Twicing or not will not incur big difference on both speed and GFLOPs. What is the reason behind that? I would appreciate a more detailed efficiency analysis & comparison in the rebuttal phase if possible." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Novelty: The authors introduce the Twicing Attention mechanism to slow down the eigenvalue decay associated with representational collapse.\n2. Theoretical Contribution: the authors provide mathematical validation for the Twicing Attention’s capability to retain information across layers.\n3. Experiments: the authors evaluate their proposed methods on both language models and vision models." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The self-attention mechanism's representational capacity diminishes significantly across layers, and this oversmoothing effect is reducing overall performance. This paper introduces Twicing Attention, a novel mechanism that connects self-attention computations with low-pass non-local means smoothing filters. By employing a kernel twicing procedure, it alleviates the low-pass effects of NLM smoothing while preserving meaningful information from residuals. Twicing Attention offers slower decay of representational capacity and improved accuracy across different data modalities. Significant performance improvement brought by Twicing attention is observed in multiple tasks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Limited improvement: The gains in clean data settings (such as on ImageNet in Tab. 1) are modest.\n2. Lack of comparison: the work does not compare its method with alternative solutions that address oversmoothing, such as regularization strategies.\n3. Lack of ablations: the authors are suggested to consider applying the proposed method at different layer depths or intervals and evaluate their difference." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 2 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. For pure curiosity, I would like to ask what the authors think the performance of this method would be in more extreme cases, which in this case refers to two main scenarios: first, the performance on LLMs with a very large number of parameters. Second, on non-classical Transformer structures, such as Linear Transformer and other analogs." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper is well-written, with clear expression of formulae and symbols, and will be highly readable.\n2. The authors discuss the recurring problem of decay of representational capacity in Transformer, which has also been recognized as a manifestation of rank collapse in other studies. Instead of directly continuing the study of related work on rank collapse, the authors start with the NLM and try to gradually restore the cause of this phenomenon and again based on the proposed method that can alleviate this problem, the research angle is more interesting and also quite theoretical significance.\n3. The author's description of the solution is complete and accompanied by thorough proofs, the process is clear and easy to understand, and the work done is very informative." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper propose the Twicing Attention, a novel attention mechanism that uses kernel twicing procedure in nonparametric regression to achieve slower decay of representational capacity and improved accuracy across various data modalities and tasks. And the design of this module builds on the study of the connection between self-attention and NLM smoothing filters. The method was tested on a public dataset, yielding promising results." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Admittedly, the authors' work is very rich and makes a very profound contribution at the theoretical level, but in my humble opinion, the authors' approach serves as a skillful level of reconciliation that moderates the rank collapse in depth, whereas a similar reconciliation skill is actually not uncommon in rank collapse-related research directions. I am not accusing the authors of not being innovative enough, but I hope that the authors can go further at the theoretical level and expand the sequence model that can really reduce the loss of information unlike the classical Transformer.\n2. The author's research is more profound, but the experiments are less adequate, with too few test datasets and too few comparison methods. I tend to think that this is the result of too much time constraints, and I hope that the author will add more datasets as well as other experiments on the Transformer if there is enough time." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "N/A" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Visualizations on earlier layers and more heads of the transformers would help to strengthen your claim.\n- Please refer to the weakness.\n- I am open to increase my score if you alleviate my concerns." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Theoretical foundation: The paper provides analysis connecting self-attention to NLM filters. \n- Fluent presentation flow: Messages of this paper are presented well, with well-demonstrated background knowledge and motivation.\n- Empirical validation: The authors provide visualizations of attention heatmaps, which validates their claim that their method preserve the diversity of token representations" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces the Twicing Attention mechanism, drawing inspiration from established connections between self-attention and low-pass non-local means (NLM) smoothing filters. The authors demonstrate two key advantages of their proposed method: 1) a theoretically proven slower decay of representational capacity across transformer layers, and 2) improved performance on both vision and language tasks across multiple datasets. The paper's primary contribution lies in its theoretical framework. It first establishes that representation collapse in transformers stems from the inherent low-pass characteristics of NLM filters. The authors then provide proof showing that the twicing formulation ($2A^2-A$) offers superior theoretical properties compared to standard attention ($A$), particularly in preserving token diversity and meaningful feature representations." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- **Narrow Problem Framing**: The paper's central premise regarding \"representation collapse\" in transformers warrants closer scrutiny. Recent research has demonstrated that this phenomenon is not inherent to transformer architectures. For instance, DINO(Caron et al., 2021) demonstrates that self-supervised training can produce well-structured, diverse token representations in Vision Transformers. Furthermore, Darcet et al. (2024) provide evidence that apparent \"collapse\" may actually reflect a more nuanced information distribution pattern, where artifacts in attention heatmaps encode global information while non-artifact tokens maintain distinct representations, albeit with lower similarity to the CLS token.\n- **Additional computational cost and marginal empirical improvements**: Performance increase in Table 4 is in trade of computational cost. Hardly can engineers be convinced to substitute the original attention with the proposed one.\n- **Limited Evaluation Scope**: The authors report the empirical performance on classification tasks for vision models. Yet dense tasks such as segmentation are more direct and effective in evaluating the structure of patch representations produced by the method." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We propose a novel attention mechanism to enhance expressive power of transformers by leveraging residual information." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024transformer,\ntitle={Transformer Meets Twicing: Harnessing Unattended Residual Information},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=16kG5aNleS},\nnote={under review}\n}" }, "abstract": { "value": "Transformer-based deep learning models have achieved state-of-the-art performance across numerous language and vision tasks. While the self-attention mechanism, a core component of transformers, has proven capable of handling complex data patterns, it has been observed that the representational capacity of the attention matrix degrades significantly across transformer layers, thereby hurting its overall performance. In this work, we leverage the connection between self-attention computations and low-pass non-local means (NLM) smoothing filters and propose the Twicing Attention, a novel attention mechanism that uses *kernel twicing procedure* in nonparametric regression to alleviate the low-pass behavior of associated NLM smoothing with compelling theoretical guarantees. This approach enables the extraction and reuse of meaningful information retained in the residuals following the imperfect smoothing operation at each layer. Our proposed method offers two key advantages over standard self-attention: 1) a provably slower decay of representational capacity and 2) improved accuracy across various data modalities and tasks. We empirically demonstrate the performance gains of our model over baseline transformers on multiple tasks and benchmarks, including image classification and language modeling, on both clean and corrupted data." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "transformers", "self-attention", "oversmoothing", "nonlocal smoothing", "nonparametric regression" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/465d7006fcae05f2c55e553fa21c9e7843e3fdd4.pdf" }, "presentation": null, "primary_area": { "value": "foundation or frontier models, including LLMs" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/feb0fa1c50098dc8e35348fc610d43ac06bcf816.zip" }, "title": { "value": "Transformer Meets Twicing: Harnessing Unattended Residual Information" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "- For the switch signals (Fig. 2 a-c, also Fig. 3A low noise setting), the inferred ODE parameter time series seem to exhibit high frequency oscillations on top of the correct switch-like dynamics. Is there an intuitive explanation why the encoder-decoder architecture struggles in inferring the correct switching dynamics and how this could be addressed?\n- Results of Fig. 3B look rather weak to me, can the authors report Pearson’s $r$ of noise lvl vs. std?\n- I’m confused by section 4.6 & Fig. 7; How exactly does the dynamic SINDy approach compare now to the proposed baseline methods based on SLDS and (vanilla?) SINDy with a group sparsity norm? I think Fig. 7 would be much clearer if the authors would find a design to compare all comparison methods side-by-side.\n- ll. 409-411: Can the authors provide references for the mentioned studies?\n- How do other methods like reservoir computing compare to the dynamic SINDy approach qualitatively and quantitatively in the settings discussed in the manuscript (see e.g. [1])?\n- How does the approach perform on e.g. benchmarks used in [2], which exhibit different bifurcations than the ones discussed in this paper?\n\nI am very happy to increase my score if the authors adequately address my concerns and questions.\n\nReferences:\n\n[1] Köglmayr, Daniel, and Christoph Räth. \"Extrapolating tipping points and simulating non-stationary dynamics of complex systems using efficient machine learning.\" Scientific Reports 14.1 (2024): 507.\n\n[2] Patel, Dhruvit, and Edward Ott. \"Using machine learning to anticipate tipping points and extrapolate to post-tipping dynamics of non-stationary dynamical systems.\" Chaos: An Interdisciplinary Journal of Nonlinear Science 33.2 (2023)." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Learning a parsimonious representation of non-autonomous DS is extremely important and relevant in many scientific disciplines, which makes the approach very promising.\n- I think it is highly interesting that the encoder-decoder architecture is able to predict the ODE parameters with this level of fidelity in an unsupervised fashion (as there is no direct reconstruction loss for the ODE parameter time series involved in the loss function (7)).\n- The method is tested against other baselines and also on a real-world dataset (C. elegans)." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes ‘dynamic sparse identification of nonlinear dynamics’ (dynamic SINDy), a deep learning framework for identifying governing equations in noisy, non-stationary and nonlinear dynamical systems (DS). By combining variational autoencoders (VAEs) and previous work on SINDy, it enables unsupervised inference of the underlying ODE systems’ parameters while extracting a global and parsimonious nonlinear dynamical model. The approach is validated on both synthetic and real-world data and is compared to other methods in the field, demonstrating great potential for scientific machine learning community." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The authors should stick to the ICLR style guide and use the 'author, year' reference style instead of mere numerical numbers (i.e. APA style instead of IEEE). This increases readability and helps the reader to understand the train of thought of the authors, as one directly sees on which work the authors base certain statements.\n- Center box in Fig. 1B is in parts hard to read as (font) sizes vary a lot. I think it would be better to shrink down Fig. 1A a touch and to increase size of Fig. 1B, especially as it describes the main framework of the manuscript.\n- I also think the figure group titles ('suptitles') above Fig. 1, 2, and 4 are superfluous and their message should be put into the figure caption. This would create additional space (e.g. to compensate for the change in referencing style).\n- I think ‘dynamic HyperSINDy’ deserves a bit more attention in the main text, which lacks explanation on how this approach really works. Explaining this method in the supplement makes the corresponding results a bit hard to read and almost forces the reader to read the supplement section 1.2.2.\n- All of the employed (benchmark) datasets are fairly low dimensional (2-3D). The authors do not address the scalability of the method to high dimensional systems (which can not be sufficiently described by the first few PCA components). I think this is a major drawback, as this setting is highly relevant to many real-world systems.\n\nMinor details:\n- typo: Fig. 3B y-axis label say “approxiate std”\n- l. 353 It just says 6A and 6B, while the authors probably reference Fig. 5A and 5B? Also l. 360 it says 6C instead of 5C.\n- typo: supplement l. 262 it says weight decay of 1e5 (I assume 1e-5?)" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "have the authors applied their method to any datasets requiring a higher dimensional latent space to see how quality of the learned dynamical system scales with dimensionality of the latent space? \n\nhave the authors considered comparisons to a method more adept at handling more smooth like transitions of dynamics such as [1] which considers a smoothly switching latent system. \n\n[1] Kurle, Richard, et al. \"Deep rao-blackwellised particle filters for time series forecasting.\" Advances in Neural Information Processing Systems 33 (2020): 15371-15382." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "dynamical system reconstruction or system identification is an important topics, and learning more interpretable models of system dynamics, such as time varying ode representations like the authors, has broad application to many fields. additionally, the authors consider a comprehensive amount of toy experiments — i appreciate that the authors considered experiments with several time varying motifs (i.e. switching, sigmoid, switch, fourier) and show that their method can recover the time varying coefficients with a calibrated measure of uncertainty." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "the authors introduce dynamic sindy — an extension of the sindy for uncovering nonstationary dynamical systems from data. the authors use a time-series VAE architecture to map their data to time-varying coefficients that are linearly combined with a fixed library of basis functions to produce an estimate of the data derivative. they conduct experiments using several toy datasets showing that their dynamic sindy can recover the coefficients of time-varying dynamical systems, even in the case that the entire system is not observed. finally, they show on low-dimensional representations of c. elegans neural recordings, that their method recovers a representative dynamical system of the first principal component." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "i found the separation between what has previously been done in the literature and what are the authors exact main novel contributions to be unclear; for example, more precise statements about the differences with hypersindy (around line 096) would have been very helpful. in its current form, it is hard to parse from the manuscript what their exact technical advances are.\n\na lot of real-estate in the paper is taken up by the experimental plots. often i found the amount of information conveyed by the plots disproportionate to the amount of space they take up — making more compact figures seems like it would work to the authors advantage. additionally, information could be conveyed better i.e. thick lines and their ordering (i.e. green/blue lines in Fig 2d are not clear). Fig 1B has a lot of small labels and the zoomed out view of timeVAE does not feel like it helps much." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Variational inference is known to provide overconfident uncertainty estimates because of the KL term encouraging mode-covering behavior. Can the authors discuss this further and the impact of this on the method?\n- How robust are the identified parameters? Is there a quantitative relationship between the robustness and (1) the size of the library, (2) the level of noise in the system?\n- It is stated that the method can only deal with non-stationarity arising from separable time-varying variables. Can the authors elaborate more on why this is the case?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The authors address an important and often overlooked problem in modeling time series data: learning interpretable models from non-stationary dynamical systems.\n- The problem and the proposed method are presented clearly, and the paper is generally easy to follow." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose DynamicSINDY, an approach that aims to learn sparse ODEs with time-varying coefficients from noisy, non-stationary time series data. The authors achieve this by combining SINDy with sequential VAEs to probabilistically infer the coefficients and their time-varying values. The method is evaluated on three synthetic datasets and a calcium imaging dataset of C. elegans." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- While the motivation of the problem is clear, I find that the conducted experiments fail to convince the reader of the method's impact in real-world settings. The experiments mainly focus on synthetic datasets that are artificially created to fit this problem and avoid most challenges often encountered in real-world datasets (high dimensionality, non-Gaussian noise, large space of possible coefficients). I think the paper would be greatly benefit if the authors can demonstrate the method in such contexts. \n\n- The C. elegans dataset used to demonstrate the method is quite simple, and the method is only applied to low-dimensional representations obtained via PCA (which in this case is enough to explain the data). The impact of using DynamicSINDY in this case is not well motivated, and the obtained results don't add any much scientific insights, especially since identifiability is not discussed.\n\n- While this is not necessarily always a weakness, the proposed method is a straightforward combination of two existing approaches. Taken together with the limited experiments section and the lack of significant technical innovations, I find the overall contribution of the paper in its current form rather limited." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- What can one do with the uncertainty? Is it necessary for accurate estimate? Would the uncertainty provide insights for scientific questions? Suggestion: elaborate more or showcase scenarios where the uncertainty is useful vs. method.\n- How does the method perform for noise driven dynamics? e.g. system with multiple meta-stable points or line attractors.\n- What scientific insights the proposed method could offer for the systems that don't have particular a priori form of ODEs? If we don't know u(t) switches, would it discover that?\n- Line 461 typo: rLDS\n- Suggestion: put Dynamics SINDy reconstruction together with SLDS result in Fig. 7." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Modeling Time-Varying Coefficients: Dynamic SINDy employs deep generative models, specifically variational autoencoders (VAEs), to learn the time-varying nature of ODE coefficients. This enables the identification of non-autonomous systems that exhibit complex dynamics.\n- Uncertainty Quantification: The use of VAEs allows dynamic SINDy to quantify uncertainty in the estimated ODE coefficients. \n- Latent Variable Discovery: Dynamic SINDy can effectively uncover hidden (latent) variables that influence the observed dynamics. This is demonstrated through an example with the Lotka-Volterra equations, where only prey population data is available.\n- Application to Real-World Data: The paper validates dynamic SINDy's capabilities on both synthetic and real-world data, i.e. the C. elegans data." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents dynamic SINDy, a machine learning method designed to identify the governing equations of noisy, nonlinear, and non-autonomous dynamical systems from data. Dynamic SINDy combines variational inference with sparse identification of nonlinear dynamics to identify time-varying coefficients in sparse ordinary differential equations. The method is particularly valuable for non-stationary systems with changing parameters over time. The contributions include\n- Modeling Time-Varying Coefficients: Dynamic SINDy employs deep generative models, specifically variational autoencoders (VAEs), to learn the time-varying nature of ODE coefficients. This enables the identification of non-autonomous systems that exhibit complex dynamics.\n- Uncertainty Quantification: The use of VAEs allows dynamic SINDy to quantify uncertainty in the estimated ODE coefficients. This is crucial for understanding the reliability of the identified model." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Uncertainty Quantification: though sec 4.2 shows the estimated standard deviation follows the truth, it only confirms the accuracy of estimation. Nonetheless, it is unclear what use the uncertainty could provide.\n- The examples are mostly using systems that driven by mean/input processes. It's unclear how the proposed method would perform for noise driven dynamics.\n- In the C. elegans example, the assumed form depends heavily on prior knowledge (dimensionality, input) on the dynamics. Though the proposed method has shown very good performance, it's unclear what scientific insights the proposed method could offer especially for the systems that people have limited knowledge about." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We introduce a method for identifying non-autonomous differential equations and discovering latent variables in dynamic systems, validated on synthetic data (e.g., the Lorenz system) and applied to neuronal activity data." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024deep,\ntitle={Deep Generative Modeling for Identification of Noisy, Non-Stationary Dynamical Systems},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1762Fbr4HK},\nnote={under review}\n}" }, "abstract": { "value": "An important challenge in many fields of science and engineering is making sense of time-dependent measurement data by recovering governing equations in the form of differential equations. We focus on finding parsimonious ordinary differential equation (ODE) models for nonlinear, noisy, and non-autonomous dynamical systems and propose a machine learning method for data-driven system identification. While many methods tackle noisy and limited data, non-stationarity – where differential equation parameters change over time – has received less attention. Our method, dynamic SINDy, combines variational inference with SINDy (sparse identification of nonlinear dynamics) to model time-varying coefficients of sparse ODEs. This framework allows for uncertainty quantification of ODE coefficients,\nexpanding on previous methods for autonomous systems. These coefficients are then interpreted as latent variables and added to the system to obtain an autonomous dynamical model. We validate our approach using synthetic data, including nonlinear oscillators and the Lorenz system, and apply it to neuronal activity data from C. elegans. Dynamic SINDy uncovers a global nonlinear model, showing it can\nhandle real, noisy, and chaotic datasets. We aim to apply our method to a wide range of problems, specifically to dynamic systems where complex parametric time dependencies are expected." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "system identification", "non-autonomous differential equations", "dynamical systems", "variational inference", "variational autoencoders", "SINDy", "sparse regression", "uncertainty quantification", "latent variable discovery", "biophysics applications", "biology", "neuroscience" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/1131bdc48d0fc37fdb392559d4c95a6c5a799148.pdf" }, "presentation": null, "primary_area": { "value": "learning on time series and dynamical systems" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/8bcd2f7a9a63c6091074f573e7b45c130491bc0c.zip" }, "title": { "value": "Deep Generative Modeling for Identification of Noisy, Non-Stationary Dynamical Systems" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See the questions raised above in the weakness." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1) The idea of multiplying Chebyshev polynomials of inputs to features as element-wise (Hadamard) product is novel.\n\n2) Experiments show that such ChebyNet of using high order (1,...,9) Chebyshev polynomials may often improve the precision over the plain networks and be often better than using ordinary polynomials (PolyNet)." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies a new approach for interactions between non-adjacent layers in neural networks. Existing interactions among non-adjacent layers are typically studied as additive shortcuts as ResNets, dense connections, and attention mechanisms. This paper aims to bring a new type of interactions between nonadjacent network layers, by multiplying Chebyshev polynomials of inputs (up to a downsampling) to features as element-wise or Hadamard products. Since the 0-order Chebyshev polynomial is the identity, such a scheme can be regarded as an extension of existing network features $f(x)$ to those multiplied element-wisely by sums of high order Chebyshev polynomials of inputs, i.e. $f(x) \\circ \\sum_{i=0}^n L_i(g(x))$, where $L_i$ is defined as the Chebyshev polynomials of the first kind recursively and $g(x)$ is a downsampling operation to align the dimensionality of input $x$ with the feature $f(x)$. \n\nThe motivation of exploiting Chebyshev polynomials roughly lies in the fact that their roots, Chebyshev nodes, actually provide a tight bound in polynomial interpolation of continuous functions that minimizes the Runge oscillation phenomenon. Moreover, Chebyshev polynomials in the first kind has a recursive representation which can be easily implemented with deep neural networks. \n\nThe utility of such a construction is demonstrated by several experiments: low dimensional numerical function approximation, MNIST image generation using UNet-diffusion, learning of some dynamical systems (2-body, 3-body, and real pendulum problem), UNet image segmentation (ACDC and BraTS19), and image classification (Cifar-10 and Cifar-100)." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1) The reproducible codes are not provided with the paper anonymously. Since it is mainly an experimental paper for ICLR, reproducible research is necessary to evaluate the experimental results. \n\n2) The motivation of designing ChebyNet architecture seems not clear enough. Among the various possibility of non-adjacent layer interactions, why do the authors choose elementwise product between the sums of Chebyshev polynomials of inputs and features? It seems to me that the 0-th order Chebyshev polynomial recovers the original networks. But what about high order polynomials? Why not additive form? Why not using weighted sum of polynomials while the weights could be tuned?\n\n3) One motivation of exploiting Chebyshev polynomials roughly lies in the fact that their roots, Chebyshev nodes, actually provide a tight bound in polynomial interpolation of continuous functions that minimizes the Runge oscillation phenomenon. Does this property lead to any particular consideration of constructing ChebyNet architecture? Moreover, why does the recursive formulation provide superior numerical stability?\n\n4) In the performance metrics, margins of improvement over the baseline are sometimes small and we are not sure if the improvements are significant. Since this is the first kind of experiments for the proposed methods, it would be better to include certain error bars to account for randomness in evaluations. \n\n5) Figure 5 shows the differences between Cheby-CNN and Poly-CNN in image classification, highlighting the negative correlations on the low orders (0-2) Cheby-CNN. The authors suggest that \"The strong correlation among high-order features suggests that low-order features are already sufficient for representing the underlying information, indicating potential for parameter compression\". However, from Table 5, middle to high order polynomials seem with high performance as well. Is there a principle in polynomial order selections? By the way, in the last row of this table, some number like 77.1, 76.8 seems missing the highlighted bold font as they are higher than the baseline." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Please refer to the weaknesses, particularly the motivation for using Chebyshev polynomials to enhance DNNs. How does this method compare to simpler residual or dense connections in terms of inter-layer interaction benefits?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1.\tChebyNet is adaptable and can be seamlessly integrated into multiple existing architectures, such as UNet and HNN, with minimal implementation complexity.\n2.\tThe approach is tested on a range of tasks and exhibits performance gains in most cases, supporting its practical efficacy.\n3.\tChebyNet shows a robust capacity for approximating various mathematical functions, with Table 6 in the appendix indicating superior performance over MLP and Poly-MLP in approximating elementary functions like sign(x) and tanh." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces ChebyNet, a novel architecture aimed at enhancing neural networks by fostering connections between non-adjacent layers, an area typically underexplored in conventional networks. The core motivation is the limited interaction between distant layers, which can constrain a network's capacity to model complex functions. ChebyNet addresses this by employing Chebyshev polynomial basis functions to augment layer connections, which are then fused with outputs, effectively enhancing the network’s representational power. The proposed method is evaluated across various tasks, including regression, image generation on MNIST, and classification, demonstrating that ChebyNet is versatile and improves performance in numerous settings." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tThe motivation for ChebyNet could be further clarified. While the paper states that existing networks lack inter-layer connections, there are established models, like DenseNet, that enhance layer interactions. Thus, the benefit of Chebyshev polynomial-based connections versus simpler dense, residual or pyramid connections remains unclear.\n2.\tFigure 1 could be refined for clarity, as it currently suggests that polynomial connections link the network’s input and output directly, whereas, according to the text part, these connections are applied within layers.\n3.\tThe method’s evaluation is limited to small-scale datasets. Testing on larger benchmarks, such as ImageNet, would provide a more compelling demonstration of its scalability.\n4.\tWhile ChebyNet is posited to improve non-adjacent layer interactions, the paper lacks strong empirical / theoretical evidence to substantiate this claim fully.\n5.\tThe paper claims that Chebyshev connections can enhance the efficiency of DNNs; however, no experiments are provided to validate this claim. To my knowledge, additional connections may introduce extra memory and I/O overhead during inference. Supplementary experiments demonstrating the efficiency benefits would strengthen the paper.\"" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "please refer to weakness." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "This paper is well-crafted, effectively illustrating the concepts and experimental results.\nTo validate the proposed method, the authors conduct extensive experiments across various tasks, including image classification and image segmentation. The outcomes of these experiments confirm the efficacy of the proposed method." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the author addresses the issue that previous neural network architectures fail to explicitly model the relationships between different layers. To mitigate this, the author introduces ChebyNet, which models the recursive and polynomial relationships between layers. To validate the proposed method, the authors conduct several experiments across various tasks. The results demonstrate that incorporating relationships between layers enhances performance and suggests a promising direction for network structural design." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Despite the demonstrated experimental improvements, I have several concerns regarding the proposed method. Firstly, could the authors provide an analysis of the memory usage, inference time, and training time of the proposed method? I am interested in determining whether it requires additional resources to train the model. Additionally, the use of MNIST and CIFAR datasets might not be sufficient to thoroughly validate the method; could the authors present results on larger datasets? Furthermore, could the authors discuss the robustness of the proposed method? While modeling the relationship between different layers may increase the capacity of the model, it could also increase the risk of overfitting." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Clear method section, with a straightforward explanation of the architectures used would go a long way in understanding the significance of the experiments.\n\nWhat was the reason behind the choice of experiments and baseline architectures? \n\nWhy are Chebishev polynomials particularly good? This is not really clear from the text. Are they a fundamental ingredient that all modern architectures should use to propel their performance further?\n\nCould you run some experiments regarding the precise type of polynomials used, or more clear ablations on where and how the polynomials were applied?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "Originality\nThis work explores an under-explored direction of research. Prior work has not carried out such an extended study.\n\nQuality\nThe method has been applied to many different tasks, trying to show the potential applications to several areas where deep learning is traditionally applied.\n\nSignificance\nSpecific problems will require specific biases, and Chebyshev polynomials are indeed an interesting way to provide useful modeling biases to the available architectures." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "Current architectures used in the field of deep learning are limited in the modeling of relationships between non-adjacent layers. Skip-connections are popular additive methods for connecting non-adjacent layers, and prior work has explored the use of polynomial functions to establish relationships between layers. In this work, Chebyshev polynomials of high order have been applied to several modern architectures and evaluated on several tasks. The experiments show that adding Chebyshev polynomials to the architectures can help improve performance slightly when compared with certain baselines." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The contribution is not clear. The first contribution claimed states that ChebyNet is introduced. However, several architectures are apparently used in the experiments, where Chebyshev polynomials are somehow applied to existing architectures to attempt improvements in their performance. Please clarify the first contribution. The second contribution claims that ChebyNet consistently surpasses existing networks, however, in the experiments it is clear that this is only true for certain hyperparameter choices (which are not clear).\n\nThe methodology is far from being clear or reproducible. The polynomials are described, but how they are applied to the network architecture is never explained in sufficient detail to allow an expert to reproduce the results obtained. No details on the architecture structure, no pseudocode of the implementation, no details of the optimizer used for each experiment (it is mentioned only for one) or the learning rate, weight decay, or other details on data augmentation, and so on.\n\nThere are writing issues with the manuscript, please read it over again and fix grammatical and typographical mistakes (e.g. Implimentation as the title of section 3.3).\n\nNumerical function approximation loss un Figure 2 shows Loss against Order. What does Order mean for an MLP? Please, again, do not place results and experiments without explaining what was done. Why is the MLP failing to fit a quadratic function? Was the error achieved exactly 0? This might not be surprising given that polynomials are part of the architecture itself, but would have been interesting a more in-depth discussion.\n\nThe FID obtained on MNIST appears very high for both the Cheby-UNet, and the baseline UNet. More details on the hyperparameters used would help understand the performance. The quality of the samples also appears qualitatively worse than a simple UNet-based implementation of diffusion available on GitHub (https://github.com/bot66/MNISTDiffusion). More details are necessary.\nThe use of FID as a metric is not sufficient. As the objective is showing the ability of the architecture to fit complex functions, the log-likelihood would have also been an important metric to display, as it more closely shows the ability of a model to fit complex functions. MNIST is not enough, and at least CIFAR10 should have been used. I would also suggest CelebA, which appears more complex but is actually quite simple compared to CIFAR10 for a generative model.\n\nThere was no discussion or acknowledgment of the limitation which comes from using models with different parameter count. From the text it is not clear whether the parameters count was kept constant, or if at least the comparison could be considered fair in all experiments." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "N/A" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "- How do you integrate the Chebyshev connections into complex architectures like ResNet or UNet? Can you provide more concrete details?\n\n- Given that Equation 4 diverges from the typical polynomial sequence, what justifies calling the method polynomial-based? Is the benefit truly coming from the polynomial structure or just from additional learned connections?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Novel Use of Polynomials: Applying Chebyshev polynomial connections for inter-layer relationships is an interesting twist that brings more flexibility to the model's structure.\n2. Versatility Across Tasks: The method shows improvements across different tasks, suggesting it has general applicability." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes ChebyNet, a neural network architecture that uses Chebyshev polynomial connections to boost the network's fitting capabilities and efficiency. The idea is to go beyond typical additive shortcuts, adding both recursive connections between adjacent layers and polynomial-based relationships between non-adjacent layers. The authors demonstrate the effectiveness of ChebyNet on various tasks, like function approximation, image classification, and semantic segmentation, showing that it often outperforms standard networks with fewer parameters." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Implementation: The implementation details are unclear. In the methodology section, Equations 3 and 4 outline the connectivity patterns between layers, but there is no specific guidance on how to apply these connections to complex architectures like UNet (as shown in Tables 1 and 4), or ResNet and MobileNet (as shown in Table 5). This raise serious problems when I want to to dive in the details of the paper. Those details are also not incorporated in the Appendix as well.\n\n2. Unclear Use of Polynomials: The method appears to focus on a recursive layer connectivity similar to Chebyshev polynomials, but it doesn't actually involve using polynomials. Equation 4 resembles Equation 3 but starts with a different initial condition, leading to entirely different sequences in the recursion.\n\n3. Computation: While the paper claims the efficiency (Line 88-89, \"fewer parameters and reduced computational overhead\"), there is no actually discussion on the real computation gain with respect to different applications. From my understanding, with increased connectivity, there is a likelihood of higher computational costs, which is why architectures like DenseNet, despite their strong performance, are not widely adopted in real-world applications. The paper does not sufficiently discuss how ChebyNet handles the potential slowdown due to the additional polynomial connections.\n\n4. Limited Baseline Comparisons: The paper mainly introduce a new type of connectivity of layers, which is more on par for ResNet and DenseNet. However, the comparisons are mostly against basic versions of popular models. Adding comparisons with more sophisticated connectivity strategies would strengthen the results and make the findings more convincing." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024chebynet,\ntitle={ChebyNet: Boosting Neural Network Fitting and Efficiency through Chebyshev Polynomial Layer Connections},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=17U3nlco2r},\nnote={under review}\n}" }, "abstract": { "value": "Traditional deep neural networks (DNNs) predominantly adhere to a similar design paradigm. Even with the incorporation of additive shortcuts, they lack explicit modeling of relationships between non-adjacent layers. Consequently, this paradigm constrains the fitting capabilities of existing DNNs. To address this issue, we propose ChebyNet, a novel network paradigm to build Chebyshev polynomial connections between general network layers. Specifically, we establish recursive relationship among adjacent layers and polynomial relationship between non-adjacent layers to construct ChebyNet, which improves representation capabilities of the network. Experimentally, we comprehensively evaluate ChebyNet on diverse tasks, including function approximation, semantic segmentation, and visual recognition. Across all these tasks, ChebyNet consistently outperforms traditional neural networks under identical training conditions, demonstrating superior efficiency and fitting properties. Our findings underscore the potential of polynomial-based layer connections to significantly enhance neural network performance, offering a promising direction for future deep learning architectures." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "DNN", "Chebyshev Polynomial" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/b20c6578ac4c4c2b94a3a4f78b642688629f7836.pdf" }, "presentation": null, "primary_area": { "value": "unsupervised, self-supervised, semi-supervised, and supervised representation learning" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "ChebyNet: Boosting Neural Network Fitting and Efficiency through Chebyshev Polynomial Layer Connections" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Should the sample complexity in the non-convex case be $\\mathcal{O}(n + \\sqrt{n/m}L^2\\varepsilon^{-2})$? Letting $m = 1$, the problem reduces to the centralized finite-sum setting, where the sample complexity should be $\\mathcal{O}(n + \\sqrt{n}\\varepsilon^{-2})$ or $\\mathcal{O}(n\\varepsilon^{-2})$, as shown in [1].\n\n2. In Table 1, is a direct comparison of convergence rates with [2] appropriate? Specifically, this paper addresses a finite-sum problem, whereas [2] deals with an online setting. Since DVRGTFW cannot be directly applied to the online problem, such a comparison may be inappropriate. The authors should at least point out the differences in settings when making these comparisons.\n\n3. Finally, there are some minor issues, such as typos. \n- The Lyapunov functions defined in L.739 use the symbols $\\Phi$ and $\\Psi$ , but in several places in the following proofs, they are written as $\\phi$ and $\\psi$ (L.994, L.1069, L.1076, L.1082, and L.1085).\n- L.818. ``fastMix'' should be ``FastMix''.\n- The paper [1] has been accepted in ICML and the reference should be updated.\n\n---\nReferences\n\n[1] Aleksandr Beznosikov, David Dobre, and Gauthier Gidel. Sarah frank-wolfe: Methods for constrained optimization with best rates and practical features. In ICML, 2024.\n\n[2] Hoang Huy Nguyen, Yan Li, and Tuo Zhao. Stochastic constrained decentralized optimization for machine learning with fewer data oracles: a gradient sliding approach. arXiv preprint arXiv:2404.02511, 2024." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper shows better theoretical convergence results compared to previous works. Specifically, by incorporating techniques such as gradient tracking and multi-consensus, it extends constrained finite-sum algorithms to the decentralized setting. The convergence of DVRGTFW is analyzed using Lyapunov functions, theoretically establishing improved sample and communication complexities, which is also validated by numerical experiments." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies the decentralized constrained finite-sum optimization problem and provides a projection-free algorithm called DVRGTFW. In the convex and non-convex cases, the sample complexities $\\mathcal{O}(n+\\sqrt{n/m}L\\varepsilon^{-1})$ and $\\mathcal{O}(\\sqrt{n/m}L^2\\varepsilon^{-2})$ are established, respectively. Numerical experiments validate the performance of the algorithm." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "While improved theoretical results are established for decentralized Frank-Wolfe method, the techniques are overall similar to existing ones." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Please see Weakness." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. This paper is well written. It is easy to follow.\n\n2. The literature review is good." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper develops a decentralized stochastic Frank-Wolfe algorithm and establishes its convergence rate for both convex and nonconvex constrained problems. The experiment demonstrates the effectiveness of the proposed algorithm." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The novelty is limited. Decentralized unconstrained optimization has been well studied. This paper tries to extend those algorithms to constrained problem, where the feasible set is bounded. However, this extension is trivial. In particular, due to the bounded feasible set, it is trivial to bound the gradient variance. Actually, the proof for frank-wolfe algorithm is much easier than the unconstrained counterpart.\n\n2. As mentioned in this paper, there are some existing decentralized Frank-wolfe algorithms for DR-submodular optimization problems. What is the difference between those algorithms and this paper? Are there any unique challenges compared to those algorithms? It would be good if the authors could discuss these critical points to show the contribution of this paper. \n\n3. FastMix is a not very common communication method. It would be good to provide some background for this method. For example, in standard gradient tracking method, it is well known that $\\bar{v}_t=\\bar{y}_t$. Does FastMix also have this property? It seems the authors directly use $\\bar{v}_t=\\bar{y}_t$ in the proof. \n\n4. It would be good to provide more details about the proof. For example, how to get the third step in Line 764? It is not very clear. \n\n5. How does the heterogeneity affect the convergence rate? \n\n6. Why does IFO not depend on the spectral gap? Any explanation?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. In table 1, when $m = 1$, we should recover the complexities in the centralized setting in the convex/non-convex setting, however, for the proposed algorithm, the reviewer does not understand why it matches the bounds given in [Beznosikov et al., 2023], for example, in the convex case the table suggests $\\tilde{\\mathcal{O}}(n + \\frac{\\sqrt{n}}{\\varepsilon})$, while [Beznosikov et al., 2023] gives $\\tilde{\\mathcal{O}}(n + \\frac{1}{\\varepsilon})$.\n\n2. What is the output of Algorithm 2 FastMix? \n\n3. Is it possible to further improve the communication complexity of the algorithm so that it matches the optimal bounds?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The author manages to combine the technique of variance reduction and gradient tracking to Frank-Wolfe algorithm in the decentralized setting, convergence analysis in both convex case and non-convex case are provided, illustrating the effectiveness of the proposed algorithm DVRGTFW.\n\n2. The proposed algorithm achieves best-known incremental first order oracle complexities both in the convex case and in the non-convex case, and near optimal communication complexity in the non-convex case.\n\n3. The paper offers numerical experiments to validate the theory presented in the paper." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the authors proposed to combine the Frank-Wolfe algorithm with variance reduction as well as gradient tracking in the decentralized setting, resulting in the algorithm DVRGTFW. Convergence analysis in the convex and non-convex case are provided with numerical experiments conducted to further support the theory provided." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Though the results are interesting, the proposed method appears to be primarily a combination of established techniques, such as variance reduction, gradient tracking, and the Frank-Wolfe algorithm. As a result, the novelty of the approach may be somewhat limited.\n\n2. If I am not mistaken, the communication complexity for DVRGTFW is not better than existing methods in the convex case given its extra dependence on $\\sqrt{mn}$ as it is demonstrated in Table 1, which is a limitation of the algorithm.\n\n3. I recommend that the authors do a thorough check of the paper as there are many typos, some of them are confusing, such examples include:\n- At line 92, ''develop communication and communication efficient'';\n- At line 114, $m = 0$;\n- At line 222, $x_0 \\in \\mathbb{R}^d$,\n- There are also some notations used without introduction in the paper.\n\n4. In some of the numerical experiments, the proposed algorithm is not better than existing algorithm for an unclear reason." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024a,\ntitle={A Computation and Communication Efficient Projection-free Algorithm for Decentralized Constrained Optimization},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=17idjbdHVW},\nnote={under review}\n}" }, "abstract": { "value": "Decentralized constrained optimization problems arise in numerous real-world applications, where a major challenge lies in the computational complexity of projecting onto complex sets, especially in large-scale systems. \nThe projection-free method, Frank-Wolfe (FW), is popular for the constrained optimization problem with complex sets due to its efficiency in tackling the projection process. \nHowever, when applying FW methods to decentralized constrained finite-sum optimization problems, previous studies provide suboptimal incremental first-order oracle (IFO) bounds in both convex and non-convex settings. \nIn this paper, we propose a stochastic algorithm named Decentralized Variance Reduction Gradient Tracking Frank-Wolfe ($\\texttt{DVRGTFW}$), which incorporates the techniques of variance reduction, gradient tracking, and multi-consensus in the FW update to obtain tight bounds. \nWe present a novel convergence analysis, diverging from previous decentralized FW methods, and demonstrating $\\tilde{\\mathcal{O}}(n+\\sqrt{\\frac{n}{m}}L\\varepsilon^{-1})$ and $\\mathcal{O}(\\sqrt{\\frac{n}{m}}L^2\\varepsilon^{-2})$ IFO complexity bounds in convex and non-convex settings, respectively. \nTo the best of our knowledge, these bounds are the best achieved in the literature to date. Besides, in the non-convex case, $\\texttt{DVRGTFW}$ achieves $\\mathcal{O}(\\frac{L^2\\varepsilon^{-2}}{\\sqrt{1-\\lambda_2(W)}})$ communication complexity which is closed to the lower bound $\\Omega(\\frac{L\\varepsilon^{-2}}{\\sqrt{1-\\lambda_2(W)}})$. \nEmpirical results validate the convergence properties of $\\texttt{DVRGTFW}$ and highlight its superior performance over other related methods." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Decentralized stochastic optimization", "variance reduction", "Frank-Wolfe method" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/5142fd7eb6fb6597fd7b673fe652db27d7cc9c04.pdf" }, "presentation": null, "primary_area": { "value": "optimization" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/169b2e472d47302502590ff2ef43ee7d3cf667df.zip" }, "title": { "value": "A Computation and Communication Efficient Projection-free Algorithm for Decentralized Constrained Optimization" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weakness" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. This paper is easy to understand.\n\n2. Experiments are conducted on three new datasets." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper builds on cluster-gcn, using chordal graph partition instead of the metis algorithm. The performance of CGST was tested on three large-scale datasets. Overall, the novelty of this paper is limited, and its performance is relatively average." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The novelty of this paper is extremely limited. The main difference from Cluster-GCN is the use of a different graph partition algorithm. Moreover, the graph partition algorithm used in this paper is not original. Additionally, the random aggregation technique mentioned in section 4.3 is also used by Cluster-GCN. The only difference is that edges between different clusters have been removed.\n2. The experimental results indicate that CGST's performance is suboptimal. Although the accuracy is sufficiently good, as a work on scalable training, the memory usage and training time performance are worse than the baselines.\n3. This paper does not discuss any work related to scalable training from 2022 to 2024.\n4. This paper contains many typos.\n5. This paper does not compare with baselines on standard datasets." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. In Algorithm 1, how to get the input clique tree? What's the complexity to construct such a tree? Considering to arbitrarily select an edge at each epoch, how to guarantee balanced partition?\n2. What are the strengths to partition a graph into balanced chordal graph over other balanced partition?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The organization is good to follow.\n2. The authors find a new potential of training large-scale graph." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper focuses on training GNNs on large graphs, and proposes to separate the whole graph into several balanced chordal graph. The authors try to maintain main information of the original graph structure." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Potential violation of double-blind policy. For the link at line 387, 'LICENSE' contains \"Copyright (c) 2024 Su ziyang\", where \"Su ziyang\" is a name implying one of the authors. Besides, this link contains the LATEX file of the submission not the code.\n2. Section 3 is about one page, but only contains some well-known knowledge.\n3. As shown in Figure 1, the authors argue that previous mini-batch methods suffer from information loss because of removed nodes and edges. However as shown in Figure 2, the proposed model also does not consider the nodes between different cliques.\n4. The baselines are too old, where the authors do not provide the citation for SAGN.\n5. As shown in Table 1, the proposed model cannot achieve the best memory usage and training time in all three datasets. Considering this paper studies large-scale training, these two metrics are very important.\n6. I strongly suggest the authors further check the writing:\n- Section 3 \"PREMILINARY\"\n- What's $\\mathcal{O}$ in Definition 2." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "Q1: Please discuss the difference between your paper and ClusterGCN.\n\nQ2: In Line 94, \"Under extensive experiments on four real-world datasets...\" , where are the fourth dataset?\n\nQ3: Section 2.3 has a title of \"GNN decoupling\", but the main text is about attention and skip-connection. How these concepts are related to GNN decoupling?\n\nQ4: In Line 359, \"We select six baselines to evaluate the performance...\", where are the sixth baseline?\n\nQ5: In Line 388, \"Codes are available at...\", there is no implementation code in this link, here is the tex source. It is normal not to provide the code during the review stage, but please do not deceive the reviewers.\n\nQ6: In Line 517, \"Case study...\", this is not case study." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "S1: The scalability of GNNs is an important research problem.\n\nS2: The format looks fine." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes CGST. CGST includes a balanced chordal graph partition module and a batch random aggregation module to improve performance on node classification tasks while maintaining main information." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "W1: Apart from the graph partition method, I don't see any difference between this paper and ClusterGCN. And the graph partition method is adopted from existing work.\n\nW2: The experiment setting is strange. The authors do not use common graph datasets. The results are also unsatisfied, as a scalable method, CGST performs poorly both in terms of memory and time.\n\nW3: Parts of this paper were written in LLM, e.g., Line 321, what is \"cluster graph spatial transformer\"?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "None" }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "See weakness" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Experiments are conducted on three new datasets, which is inconsistent with previous work. Testing on new datasets is commendable.\n\n2. This paper is easy to understand." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a GNN training technique based on subgraph sampling, which is based on chordal subgraph partition. The authors tested the performance of CGST training on GCN across three large graphs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. In the introduction, the figure 1 is inappropriate. Methods like CLUSTER-GCN and GAS use METIS to partition graphs, which does not result in some nodes being removed and unable to appear in the training batches.\n2. The author frequently mentions that chordal subgraph partition is a major contribution, but I notice that the work in section 4.2 originates from [1]. This significantly undermines the novelty and amount of work in this paper. The author should provide an accurate explanation and description of this.\n3. There are significant problems in the experimental section of the paper, which completely fails to meet the acceptance standards of ICLR. The author should provide experimental results for a variety of GNNs, not just limit to GCN. In terms of experimental results, CGST is also not ideal in terms of Mem usage and training time. Moreover, the author should provide experimental results for commonly used datasets, such as Products.\n\nOverall, I think this paper has significant deficiencies, especially in the experimental section.\n\n[1] Jungho Ahn,Lars Jaffke,O-joung Kwon, and Paloma TLima. Well-partitionedchordalgraphs. Discrete Mathematics." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024chordal,\ntitle={Chordal Graph Sampling-Based Mini-batch Training Algorithm for Large Graphs},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1959usnw3Z},\nnote={under review}\n}" }, "abstract": { "value": "Graph Neural Networks (GNNs) are powerful models for learning representations of attributed graphs. To scale GNNs to large graphs, many methods use various techniques, such as sampling and decoupling, to alleviate the “neighbor explosion” problem during mini-batch training. However, these sampling-based mini-batch training methods often suffer from greater information loss than decoupling-based methods or full-batch GCNs. Besides, most original segmentation methods for large graphs usually lose a large number of edges, resulting in suboptimal performance when performing mini-batch training. Therefore, we propose a Chordal Graph Sampling-based mini-batch Training algorithm for GNNs on large-scale graph datasets, called CGST. CGST includes a balanced chordal graph partition module and a batch random aggregation module to improve performance on node classification tasks while maintaining main information of the original graph structure. Experiments on three large-scale graph datasets prove the effectiveness of CGST." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Large scale dataset", "Graph neural networks" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/40f2d0d451a620b7f727b6f81b26953617033ea3.pdf" }, "presentation": null, "primary_area": { "value": "learning on graphs and other geometries & topologies" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Chordal Graph Sampling-Based Mini-batch Training Algorithm for Large Graphs" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "The anonymity of the authors is compromised, as this paper is available on arXiv at https://arxiv.org/abs/2407.05259." }, "flag_for_ethics_review": { "value": [ "Yes, Other reasons (please specify below)" ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Could the authors provide more detailed explanations regarding the choice and role of each loss term in Equation 18 and explain how they determined the relative weighting (λ, ν, α values) between the terms.\n2. Could the authors provide a comparison of training time and the number of training parameters for MSCGM versus other models?\n3. Could the authors to provide a detailed algorithm or pseudocode for MSCGM training, similar to what they provided for BBDP Algorithm." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. MSCGM’s wavelet-based decomposition and conditional modeling shows substantial improvements in sampling speed and better reconstruction quality.\n2. By adapting the generative approach to frequency characteristics, MSCGM enhances detail in restored images, especially in high-frequency components crucial for microscopy images.\n3. The authors presented a new loss function." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a multi-scale conditional generative model (MSCGM) aimed at enhancing microscopic image restoration by combining wavelet transforms and a Brownian Bridge diffusion process. The authors leverage multi-scale wavelet transforms to efficiently model low- and high-frequency image components, significantly improving the generation quality and speed of image restoration compared to traditional diffusion models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Equation 18 combines multiple objectives—L2 loss, Structural Similarity Index Measure (SSIM), and Wasserstein distance—but the rationale behind each component’s inclusion is not fully explained. Additionally, the roles and relative importance of the scaling parameters λ, ν, and α are unclear. \n2. The training procedure for MSCGM is not explicitly described. Unlike the clear training steps outlined for BBDP, MSCGM lacks a step-by-step description of its training pipeline. \n3. While Table 1 compares MSCGM with other models in terms of PSNR, SSIM, and sampling time, it does not include training time or the number of trainable parameters for each method. Without these metrics, it is challenging to gauge MSCGM’s overall computational cost relative to other approaches. Including such details would provide a more comprehensive view of the model’s efficiency.\n4. In Section 4.2, the authors state that FID is considered as an evaluation metric. However, this metric is not included in Table 1. As FID is widely used in assessing generative models for image quality, its inclusion would offer further insights into MSCGM’s performance in distributional similarity to real images.\n5. Equations from 4 to 15 are borrowed from BBDP paper. It is better to include them under the Preliminaries section." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "Please see my concerns in Weakness" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The authors recognize the loss of detail in LDM, a known issue, and apply it to the microscopic image restoration context, an interesting direction.\n2. They introduce the novel idea that the Brownian bridge stochastic process could effectively integrate conditional images." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose a multi-scale conditional generative model (MSCGM) for image restoration, incorporating multi-scale wavelet transforms and a Brownian bridge stochastic process. The wavelet transform is included due to its reversibility, which maintains information integrity in the latent diffusion space, in contrast to traditional Latent Diffusion Models (LDM). The Brownian bridge stochastic process is leveraged to introduce conditional images in both forward and reverse processes. While the authors aim to address microscopic image restoration, the motivation and results in the paper do not consistently support this focus." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. **Lack of Consistency:** The paper lacks organization and clarity. Although the title emphasizes \"Microscopic Image Restoration,\" the experiments primarily focus on \"Natural Image Super-resolution\" and \"Low-light Natural Image Enhancement.\" Only a small subset of results explores microscopic images. If the model is intended for general image restoration, it would be more accurate to propose it as a ‘unified image restoration’ model. I suggest the authors either refocus their experiments more heavily on microscopic image restoration to align with the title, or broaden the title to reflect the wider scope of image restoration tasks covered in the paper.\n \n2. **Introduction Needs Refinement:** The introduction lacks a clear problem definition and research motivation. The first two paragraphs provide a broad overview of diffusion processes that diverges from the paper’s focus. The discussion on latent diffusion downsampling is a well-known issue and could be alleviated by higher resolutions. The authors should clearly articulate why microscopic images especially require the multi-scale wavelet transform in the introduction. Please include a discussion of how their approach compares to or builds upon these existing wavelet-based diffusion models in the Introduction, highlighting any key differences or improvements.\n \n3. **Lack of Acknowledgment of Prior Work:** The paper does not credit previous studies applying wavelet transforms in diffusion models, which could mislead readers into believing the concept originated here. Papers like \"Wavelet Diffusion Models are Fast and Scalable Image Generators (CVPR 2023)\" and \"Training Generative Image Super-Resolution Models by Wavelet-Domain Losses Enables Better Control of Artifacts (CVPR 2024)\" are directly related and should be cited with comparisons to clarify this study’s contributions.\n \n4. **Figure 1 Illustration Issues:** The paper title focuses on \"Microscopic Image Restoration,\" yet Figure 1 uses natural images. Including examples of microscopic images to show the degradations introduced by LDM and Refusion compared to MSCGM would enhance clarity.\n \n5. **Methodology Development Clarity:** The description of the wavelet transform on page 4 is overly general, with key details moved to the appendix. Clear explanations of any novel model designs or algorithmic adaptations should be provided in the main text.\n \n6. **Quality of Mathematical Presentation:** Symbols in the equations are used without proper declarations or explanations. Inconsistent symbols, like the variable for the normal distribution \\( N \\), further detract from clarity.\n \n7. **Algorithm 1 Lack of Context:** Algorithm 1 on page 5 is underdeveloped. Symbols are not defined before use, and the algorithm lacks defined input-output requirements.\n \n8. **Figure 2 Diagram Confusion:** Figure 2 is difficult to interpret. The illustration doesn’t clearly label network modules, workflow processes, or shared parameters (only a line is shown), which fails to clarify the model structure effectively.\n \n9. **Lack of Dataset Information:** The results section includes evaluations of microscopic images, but there’s no description of the dataset. Is it public or private? What is the image count? Without these details, readers cannot analyze or reproduce the results. Please provide a detailed description of the microscopic image dataset used, including its source, size, and any preprocessing steps applied.\n \n10. **Insufficient Ablation Studies:** Results provide only a simple comparison with LDM, without deeper exploration of MSCGM’s components or ablation studies to justify the performance benefits of each module.\n \n11. **Unconvincing Model Performance:** The model’s performance requires further validation through comparison with advanced models. Numerous diffusion-based image restoration models from 2024 exist, yet none are used for comparison. This weakens the paper’s credibility. Key diffusion-based image restoration works worth considering include: \n - RDDM ([link](https://cvpr.thecvf.com/virtual/2024/poster/31373)) \n - HIR-Diff ([link](https://cvpr.thecvf.com/virtual/2024/poster/29665)) \n - WF-Diff ([link](https://cvpr.thecvf.com/virtual/2024/poster/30059)) \n - DeqIR ([link](https://cvpr.thecvf.com/virtual/2024/poster/31759)) \n - GDP ([link](https://cvpr.thecvf.com/virtual/2023/poster/22095))" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Especially considering that inference time is one of the main benefits, why was it not compared to models with fewer step counts or at least an in-depth analysis of how step counts influence the SOTA model performance? E.g. [2], or other methods that can be applied to the problem domain?\n\n[2] Phung, Hao, Quan Dao, and Anh Tran. \"Wavelet diffusion models are fast and scalable image generators.\" Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "• Integrating the Brownian Bridge Diffusion Process (BBDP) with adversarial learning in a multi-scale wavelet diffusion framework is innovative, enhancing image quality and stability. \n\n• The model achieves notable speed improvements, delivering faster inference without sacrificing image quality. \n\n• Performance remains consistent across diverse experiments, demonstrating robustness on both microscopy and natural images." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors present a novel multi-scale generative model that leverages the Brownian Bridge process within the wavelet domain. This approach enhances inference speed while maintaining high image quality and diversity. The model is further integrated with computational microscopy workflows, expanding its applicability. The authors evaluate its performance on both microscopy and natural image datasets, demonstrating that it achieves slightly better results compared to existing methods such as IR_SDE, Refusion, and BBDM, with the added advantage of faster inference." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "• The paper lacks a clear motivation for applying this model to computational microscopy workflows. The rationale for this specific application is unclear and lacks context, the relevance to microscopy appears out of place. A discussion on how this functionality benefits microscopy would help justify this direction and clarify its practical utility.\n\n• The primary advantage of this method is its reduced inference time; however, the paper lacks a direct comparison with other methods that similarly aim to improve efficiency. Including such a comparison would provide valuable context and help quantify the benefits more clearly.\n\n• The general evaluation lacks depth and is missing ablation studies. \n\n• There appear to be configuration issues with the comparison methods. For instance, IR-SDE [1] is cited as requiring 100 steps, but the authors use 1000, which significantly prolongs inference time. With the correct configuration (100 steps), the inference time should drop from 32 seconds to approximately 3 seconds.\n\n• The choice of metrics is limited and somewhat inadequate for a super-resolution task. Relying solely on PSNR and SSIM may overlook important aspects of image quality. Including pixel-based metrics would provide a more comprehensive evaluation and might show shortcomings of the proposed method.\n\n\n[1] Luo, Ziwei, et al. \"Image restoration with mean-reverting stochastic differential equations.\" arXiv preprint arXiv:2301.11699 (2023)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "The paper demonstrated that the low-frequency coefficients in higher scales show Gaussian tendency and thus applied this to BBDP. The idea is novel and well hypothesized, but it would be helpful if other DM methods, such as IR-SDE and ReFusion methods that are implemented on 4x super-resolution experiment, are also tested on microscopy image dataset. Only CMSR (GAN: non-diffusion model), is compared at the moment, not showing the effectiveness of proposed near-Gaussianity assumption.\nSimilarly, applying BBDM to full resolution image does not seem to be fair comparison. Since many works demonstrated the effectiveness of multi-scale diffusion models, BBDM should be implemented in a same manner as the proposed method to prove the superiority of WT instead of other compression technique. Please conduct an ablation study that replaces WT with simple down-sampling.\nIs there any specific reason why the proposed work adopted BBDM which was initially designed for image translation where input and target domains are different? Super-resolution tasks seem to have similar domains for input and target. Justify the choice of BBDM for super-resolution." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper analyzed the characteristics of microscopic images and proposed adequate methodology to address the sparsity and non-Gaussianity. Since the wavelet transformation divides the image into two subbands (high- and low- frequency coefficients) losslessly, handling each subband in a different manner is original. \nThe contribution of the work is clear and well demonstrated. \nIn addition, the work could be further applied to different modality images where sparse or non-Gaussianity exist." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposed a multi-scaled generative model that uses a diffusion model (DM) for low-frequency image and a GAN for high frequency images. The wavelet transform provides multi-scale images without lossy encoding process. The lossless compression is particularly important for microscopic imaging where high-frequency component are sparse and non-Gaussian. Additionally, the authors showed the near-Gaussian property of low-frequency component and thus employed Brownian Bridge Diffusion Process (BBDP). The idea of employing different networks (DM and GAN) to different resolutions according to the characteristics of microscopic dataset is novel. The proposed MSCGM (multi-scale conditional generative model) showed improved super-resolution result with fast inference time." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Although the idea of the paper is novel, the effectiveness of the work has not been thoroughly assessed. The use of WT and the superiority of the proposed method compared to conventional method should be further evaluated. The specific comments are described in Questions." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024multiscale,\ntitle={Multi-scale Conditional Generative Modeling for Microscopic Image Restoration},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=19QWQSsbOA},\nnote={under review}\n}" }, "abstract": { "value": "The advance of diffusion-based generative models in recent years has revolutionized state-of-the-art (SOTA) techniques in a wide variety of image analysis and synthesis tasks, whereas their adaptation on image restoration, particularly within computational microscopy remains theoretically and empirically underexplored. In this research, we introduce a multi-scale generative model that enhances conditional image restoration through a novel exploitation of the Brownian Bridge process within wavelet domain. By initiating the Brownian Bridge diffusion process specifically at the lowest-frequency subband and applying generative adversarial networks at subsequent multi-scale high-frequency subbands in the wavelet domain, our method provides significant acceleration during training and sampling while sustaining a high image generation quality and diversity on par with SOTA diffusion models. Experimental results on various computational microscopy and imaging tasks confirm our method's robust performance and its considerable reduction in its sampling steps and time. This pioneering technique offers an efficient image restoration framework that harmonizes efficiency with quality, signifying a major stride in incorporating cutting-edge generative models into computational microscopy workflows." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Microscopic Image Restoration", "Generative Model" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/ac6a5e380b3c32ce287cea704ff46dbafc11bd8f.pdf" }, "presentation": null, "primary_area": { "value": "generative models" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/6167139b18178ef2b1f5d714b6a6d63a86c51a1b.zip" }, "title": { "value": "Multi-scale Conditional Generative Modeling for Microscopic Image Restoration" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Please refer to the cons part. I will moderately raise my score if the authors can provide further experimental results and answer my questions." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The paper is well-written, which is quite easy to follow.\n2. The author sufficiently demonstrates that modality gap can cause dimensional collapse, leading to suboptimal performance." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the author mainly propose to solve the problem of dimensional collapse caused by modality gap in cross-modal knowlegde distillation task. Firstly, the author demonstrates the impact of modality gap on cross-modal features theoretically and empirically. To combat with this issue, a Cross-modal Information Bottleneck Approximation (CIBA) framework is proposed that extracts modality-general features through a bottleneck structure, meanwhile aligning teacher and student features with an additional loss. Experiments on several datasets demonstrates the performance of CIBA." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. My main concern is about the generalizability of the method. As mentioned in problem statements and limitations, the theorem is established on linear extractor, which is inconsistent with practical applications where non-linear encoders are widely applied. Under such conditions, the proposed concept of modality-general and modality-specific parts could be vague since the better capacity of the encoders. I can truly understand the difficulty of theorical provement with non-linear extractors, while the direct application of CIBA seems to be accessible. Can you provide further results of CIBA compared to SOTAs with more powerful encoders on current datasets (RAVDESS and MM-IMDB) to prove the superiority?\n2. In the method part, a bottleneck structure is utilized to capture mutual modality information. From my point of view, the dimension of the bottleneck feature may be a crucial parameter affecting the granularity of the extracted information. Performance seems to be fluctuant with chaning values of the param according to Fig.5(c). Can you provide more ablation on this param on more datasets? How do you choose the best bottleneck dimension?\n3. The author mainly focus on the introduction and demonstration of modality gap's impact on dimensional collapse, while the introduction of method seems to be ordinary and unremarkable. Besides, since the information bottleneck structure was proposed by earlier research, and the proposed loss is a direct combination of generation loss and KL loss, the novelty of the paper is somehow limited." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "How does the performance of CIBA compare when the assumption of orthogonality between modality-general and modality-specific features is relaxed? \n\nHow does CIBA differ from the previous CMKD approaches?\n\nHow sensitive is the CIBA framework to the choice of hyperparameters, particularly the dimension of the bottleneck feature (H)?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "The paper proposes a novel cross-modal knowledge distillation (CMKD) method by focusing on the issue of dimensional collapse in cross-modal feature distillation. The concept of modality gap and its impact on the efficacy of feature distillation is a fresh approach to understanding the limitations of CMKD. The proposal of the Cross-modal Information Bottleneck Approximation (CIBA) scheme is creative and addresses a significant problem in transferring knowledge across different modalities.\n\nThe paper is well-written. The figures and tables are clear and effectively support the textual content." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper titled \"Understanding Dimensional Collapse in Cross-Modal Feature Distillation\" investigates the challenges of transferring knowledge across different modalities in multi-modal neural networks, specifically focusing on the problem of dimensional collapse in cross-modal feature distillation (CMFD). The authors hypothesize that the modality gap between the teacher and student models leads to dimensional collapse in the student's feature space, which degrades the quality of knowledge distillation. To address this, they propose a novel framework called Cross-modal Information Bottleneck Approximation (CIBA), which aims to extract and transfer modality-general features from the teacher model to sub-dimensions of the student model's features. The paper empirically demonstrates that CIBA effectively reduces dimensional collapse and improves performance on various real-world multi-modal datasets, including RAVDESS (Audio-Image), MM-IMDB (Image-Text), and nuScenes (LiDAR-Camera). The key contributions of the paper are the theoretical and empirical investigation of the modality gap's impact on CMKD, the proposal of the CIBA framework, and the validation of its effectiveness across different modalities." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The contribution is incremental.\n\nI feel the information bottleneck approximation idea has been used extensively.\n\nWhile the proposed method is shown to outperform baseline approaches, it is unclear how it compares to the most recent and advanced techniques in the field, i.e., DML[1], DKD[2], DIST[3], C2KD[4].\n\n[1] Ying Zhang, Tao Xiang, Timothy M. Hospedales, and Huchuan Lu. Deep mutual learning. In CVPR, 2018. 1, 3, 6, 7.\n\n[2] Borui Zhao, Quan Cui, Renjie Song, Yiyu Qiu, and Jiajun Liang. Decoupled knowledge distillation. In CVPR, 2022. 1, 2, 3, 4, 6, 7.\n\n[3] Tao Huang, Shan You, Fei Wang, Chen Qian, and Chang Xu. Knowledge distillation from a stronger teacher. In NeurIPS, 2022. 1, 2, 6, 7, 8.\n\n[4] Huo F, Xu W, Guo J, et al. C2KD: Bridging the Modality Gap for Cross-Modal Knowledge Distillation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 16006-16015." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "N/A" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The theoretical and empirical investigation of the \"modality gap\"—the distributional shifts between different modalities—and its detrimental effect on CMKD, specifically leading to dimensional collapse in the student model’s feature space.\n\n2. CIBA extracts modality-general features from the teacher model and transfers them to sub-dimensions of the student’s features. This method mitigates the dimensional collapse, ensuring more robust and effective knowledge transfer." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper tries to address the challenges associated with deploying multi-modal neural networks in real-world applications, specifically focusing on the constraints of limited computing resources and complex sensor configurations. The authors explore Cross-Modal Knowledge Distillation (CMKD) as a solution for transferring knowledge from a pretrained teacher model to a more deployable student model tailored to a target modality. Despite the advancements in CMKD across various domains, the paper identifies a gap in understanding how distributional shifts between modalities—referred to as the modality gap—affect the efficacy of feature distillation. The study hypothesizes and empirically demonstrates that a significant modality gap leads to dimensional collapse within the student model's feature space, undermining performance. To mitigate this issue, the authors introduce the Cross-modal Information Bottleneck Approximation (CIBA) scheme, designed to extract and transfer modality-general features from the teacher model effectively. Experimental results on diverse real-world multi-modal datasets confirm that the proposed CIBA method successfully reduces dimensional collapse in the student model, resulting in enhanced performance. This work contributes a deeper understanding of the interplay between modality gaps and knowledge transfer in CMKD, offering a practical solution to improve the deployment of multi-modal neural networks under resource constraints." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Since RAVDESS is a relatively small size dataset. Do you try to work on VGGSound?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See the weaknesses." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1.\tThe authors successfully propose and validate that the modality gap between the teacher and student models can lead to dimensional collapse in the student’s feature space.\n\n2.\tA novel Cross-modal Information Bottleneck Approximation (CIBA) scheme is introduced to extract and transfer modality-general features from the teacher model.\n\n3.\tExperimental results across various loss functions and tasks provide strong evidence for the effectiveness of the proposed method." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper investigates the relationship between distributional shifts across modalities and their impact on the effectiveness of cross-modal knowledge distillation (CMKD), specifically addressing the issue of cross-modal feature distillation. The authors hypothesize and validate that the modality gap between the teacher and student models may lead to dimensional collapse in the student’s feature space. To address this, they propose a Cross-modal Information Bottleneck Approximation (CIBA) scheme aimed at extracting and transferring modality-general features from the teacher model. Experimental results demonstrate that the proposed distillation strategy effectively mitigates dimensional collapse in the student model." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tThe work is predicated on the assumption of linear feature extractors; however, in practical applications, most feature extractors are non-linear.\n\n2.\tIn the MM-IMDB dataset, the observed improvement is marginal. Could you please provide a more detailed explanation for this finding?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "No ethics review needed." }, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. How is the Figure 1 formulated? The manuscript does not mention the details. I think it is important for the motivation of modality-general and modality-specific knowledge analysis.\n2. How about directly apply unimodal knowledge distillation on crossmodal knowledge distillation? Could the proposed method be integrated into SOTA methds?\n\n[r1] Zhang L, Shi Y, Shi Z, et al. Task-oriented feature distillation[J]. Advances in Neural Information Processing Systems, 2020, 33: 14759-14771. \n\n[r2] Huang T, You S, Wang F, et al. Knowledge distillation from a stronger teacher[J]. Advances in Neural Information Processing Systems, 2022, 35: 33716-33727." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Research about cross-modal knowledge distillation (CMKD) on the feature view is an important topic for multimodal learning and knowledge distillation. This paper analyses the dimensional collapse induced by modality gap and propose Cross-modal Information Bottleneck Approximation (CIBA) to disentangle the general and specific knowledge, which is novel and practical.\n2. Utilizing the Mean Squared Error (MSE) loss for feature distillation (FD) is reasonable and suitable for the subsequent theoretical analysis.\n3. This work is a good extension of the modality focusing hypothesis, and gives a solid analysis and detailed solutions.\n4. This work is well written and organized. Extensive experiments on Audio-Image, Image-Text, and LiDAR-Camera crossmodal transfer are conducted." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper investigates the problem of dimensional collapse in cross-modal feature distillation (CMFD), where a student model trained on one modality aims to mimic the feature representations of a teacher model trained on a different modality. The authors hypothesize that the distributional shift, or \"modality gap\", between the teacher and student modalities leads to the student's feature space collapsing to only capture the modality-general features, resulting in suboptimal distillation performance. To address this issue, the authors provide in-depth analysis on how distributional shifts across different modalities and propose a Cross-modal Information Bottleneck Approximation (CIBA) scheme that extracts and transfers the modality-general features from the teacher to the student, allowing the student to effectively span both modality-general and modality-specific feature spaces." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Modality gap is widely studies in multimodal learning, and this paper does not give a review of previous modality gap analysis. Moreover, the cross-modal knowledge distillation on logit-level method [r1] is not mentioned and analysed.\n[r1] Huo F, Xu W, Guo J, et al. C2KD: Bridging the Modality Gap for Cross-Modal Knowledge Distillation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024: 16006-16015.\n2. The RAVDESS, MM-IMDB, and nuScenes have limited class categories. Large-scale experiments like conducting experiments on VGG-Sound (or subset) will make the paper more convincing.\n3. Related works about the 'Cross-modal knowledge distillation' are somewhat out-of-date, only one paper published in 2023 is mentioned.\n4. The proposed method is somewhat similar to online distillation [r1] and task-oriented feature distillation [r2]. How about the performance of directly employing task-oriented feature distillation [r2] on cross-modal feature distillation?\n[r2]Zhang L, Shi Y, Shi Z, et al. Task-oriented feature distillation[J]. Advances in Neural Information Processing Systems, 2020, 33: 14759-14771." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We investigate the distributional shifts across different modalities that ultimately lead to dimensional collapse in cross-modal knowledge distillation, then propose a methodology to address it." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024understanding,\ntitle={Understanding Dimensional Collapse in Cross-Modal Feature Distillation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=19ufhreGTj},\nnote={under review}\n}" }, "abstract": { "value": "To overcome limited computing resources and the complexity of sensor configurations in deploying multi-modal neural networks in real-world applications, cross-modal knowledge distillation (CMKD) aims to transfer valuable information from a pretrained teacher model to a deployable student model with the target modality. Despite the successful applications of CMKD in various fields, our understanding of knowledge transfer across different modalities remains insufficient to fully explain the efficacy of feature distillation. In this work, we investigate the relationship between the distributional shifts across modalities, referred to as the modality gap, and its impact on the effectiveness of CMKD, particularly focusing on the problem of cross-modal feature distillation. We first hypothesize and empirically validate that the modality gap between the teacher and student causes dimensional collapse in the student's feature space. To prevent such inefficiency, we propose a Cross-modal Information Bottleneck Approximation (CIBA) scheme aimed at extracting and transferring modality-general features from the teacher model. Lastly, we experimentally demonstrate that our distillation strategy effectively reduces the dimensional collapse in the student model, thereby achieving improved performance for various real-world multi-modal datasets." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "knowledge distillation", "feature distillation", "cross-modal learning", "dimensional collapse" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/b8809095347811a1d25d10c0d3171e5f6bc8a055.pdf" }, "presentation": null, "primary_area": { "value": "learning theory" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Understanding Dimensional Collapse in Cross-Modal Feature Distillation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "In the introduction, it is mentioned that BC+RL can't enable an RL agent tot acquire higher level abstractions and understanding of the RL task. This is not only very loosely defined, but likely not true. Do the authors mean that an RL agent wouldn't acquire an understanding that can be translated in language? This is very different than the claims being made.\n\nWhy use GPT-4o for generating strategies? How does Llama 3.1 compare? It would be much more preferable to have it be the same family of models.\n\nThe word \"neuro-symbolic\" is used to characterize the method, but is it really a neuro-symbolic method? To me it just seems like the neural network is additionally conditioned on language. This qualifier seems a bit of stretch.\n\n[1] Deep Reinforcement Learning at the Edge of the Statistical Precipice, Agrawal et al., 2022\n\n[2] Deep reinforcement learning that matter, Henderson et al., 2018\n\n[3] Code as Reward: Empowering Reinforcement Learning with VLMs, Venuto et al., 2024" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "The paper identifies an important area for research, that is, how to combine expert data and reinforcement learning. The proposed approach is different from some of the more traditional ways of leveraging both kinds of data, building on the strengths of LLMs to devise high level strategies." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes a way to leverage expert data for behavior cloning and RL to craft policies that are conditioned on a set of strategies which hopefully encode generalizable behavior. The authors claim that existing methods on BC+RL suffer from important issues (poor generalization, sample efficiency and interpretability), which the proposed approach can address. In particular, the authors train an open source LLM on expert data through behavior cloning by conditioning the policy on strategies that are devised by a teacher LLM (GPT-4o). The model is then trained with RL data, leading to a new list of strategies, which is then used to further guide the agent. The strategies are selected by verifying whether they help the model achieve higher performance. On a set of four environments, the paper shows that the proposed approach improves upon previous baselines." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The empirical setup brings a lot of questions. A major red flag is that there are no error bars at all and no mention of the number of seeds. Please see the rich literature on the subject of statistical relevance in decision making [1, 2]. For the tasks themselves, it is not clear why some choices are made. For example, is the max_steps=60 the default number? In the codebase of Minigrid I can see that the default value is set of 100, so an explanation would be necessary. \n\nAnother important area of doubt is concerning the strategy for updating the list of strategies. Currently, this is a complicated method that relies on evolving the strategy list with respect to performance. Why is such a complex method used? How sensitive is it to the different hyperparameters? How does it compare to simply asking GPT-4o for a new list of strategies? These are key questions that are completely unanswered.\n\nThe authors claim that generalization is a limitation of BC+RL, yet the paper does not show any experiments on generalization. This would be a great opportunity for the authors to show the compositionally that is afforded by language. It would also be a great opportunity to address another important are of concern: how much does the list of strategies affect the model? How much can you steer its behavior by changing the list? At the moment, it really isn't clear that the RL agent really responds to this conditioning.\n\nThe performance numbers reported for some of these tasks seems very low, which also comes from a limited set of baselines. In particular, I would really like to see the performance of GPT-4o on these tasks. Another family of baselines would be to compare to LLMs generating goals for the RL agent [3], which is relatively close to the ideas presented here. Notice that in that paper the results are significantly better than the numbers presented here." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weakness." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The paper is well-written and easy to follow.\n2. The authors provide illustrations of their method, which makes it clear to how it works." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposed DYSTIL which integrates LLMs into a strategy-based neuro-symbolic reinforcement learning framework. The method aims to address the generalization issue of RL." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. My biggest concern is the limited novelty and experiments. There are many papers that proposed strategy-generating methods as a summary or reflection of the past trajectories, such as [1]. The authors failed to discuss the similarities and differences between their method and these works.\n2. The experiments are only conducted in several environments from Minigrid. Whether this approach can generalize and how to design each component for different tasks remains unclear. Besides, the compared baselines are limited. I strongly encourage the authors to do literature reviews and add more baselines such as [1].\n\n[1] Shinn et al., Reflexion: Language Agents with Verbal Reinforcement Learning." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Can DYSTIL generalize to other language-based decision-making tasks, such as those solved by ReAct (e.g., ALFWorld)? How could you extend your framework to accommodate these tasks?\n2. In the GLAM baseline paper[1], the average success rate converges and reaches high performance (i.e., over 80%) at approximately 1e6 steps. Is there a reason you chose 1e5 steps for evaluation? What causes the discrepancy between your configuration and results compared to theirs?\n3. In the ablation study, dynamic strategy updates are removed, so there is no $\\mathcal{L}_2$ in the static strategy settings. Does this result in more iterations compared to the proposed method based on the same training frames? I also want to confirm whether $\\mathcal{L}, \\mathcal{L}_1, \\mathcal{L}_2$'s executions are all counted in training frames.\n4. Can the strategies generalize to novel tasks? For instance, would training on Unlock Pickup help in solving Key Corridor?\n\n[1] Carta et. al. \"Grounding large language models in interactive environments with online reinforcement learning\". In Proceedings of ICLR 2023." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- Quality: The paper presents ideas through clear text and figures, aiding understanding of the overall concepts.\n- Significance: This paper demonstrates that the proposed method outperforms two baselines in grid-world environments." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces DYSTIL, a neuro-symbolic reinforcement learning framework that integrates large language models (LLMs) to dynamically induce and internalize textual strategies, enhancing policy generalization and sample efficiency. By leveraging LLMs to provide strategy guidance based on expert demonstrations, DYSTIL improves interpretability and transparency in reinforcement learning tasks. Empirical results demonstrate that DYSTIL outperforms state-of-the-art methods by 17.75% in success rate across complex environments like Minigrid and BabyAI." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Scalability: DYSTIL’s reliance on closed-source, SOTA LLMs (e.g., GPT-4o) raises issues of scalability, reproducibility, and accessibility, especially for the model which needs to recurrently call strategy-generating LLM for each iteration. The paper also lacks ablation studies using different LLMs, which would help clarify the flexibility of using other LLMs for this work." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024dystil,\ntitle={{DYSTIL}: Dynamic Strategy Induction with Large Language Models for Reinforcement Learning},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1ABhAZCoGr},\nnote={under review}\n}" }, "abstract": { "value": "Reinforcement learning from expert demonstrations has long remained a challenging research problem, and existing methods resorting to behavioral cloning plus further RL training often suffer from poor generalization, low sample efficiency, and poor model interpretability. Inspired by the strong reasoning abilities of large language models (LLMs), we propose a novel strategy-based neuro-symbolic reinforcement learning framework integrated with LLMs called DYnamic STrategy Induction with Llms for reinforcement learning (DYSTIL) to overcome these limitations. DYSTIL dynamically queries a strategy-generating LLM to induce textual strategies based on advantage estimations and expert demonstrations, and gradually internalizes induced strategies into the RL agent through policy optimization to improve its performance through boosting policy generalization and enhancing sample efficiency. It also provides a direct textual channel to observe and interpret the evolution of the policy's underlying strategies during training. We test DYSTIL over challenging RL environments from Minigrid and BabyAI, and empirically demonstrate that DYSTIL significantly outperforms state-of-the-art baseline methods by 17.75% success rate on average while also enjoying higher sample efficiency during the learning process." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Neurosymbolic Systems", "Reinforcement Learning", "Large Language Models", "Strategy" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/d49d8dbae375e432284685600bc8e5372389f44f.pdf" }, "presentation": null, "primary_area": { "value": "neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "DYSTIL: Dynamic Strategy Induction with Large Language Models for Reinforcement Learning" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "What are the main differences between this work and that of Zhang et al. (2023)? (see Weaknesses section)" }, "rating": { "value": 1 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper proposes a heuristic to estimate which tokens contribute the most to the overall watermark signal and removes the watermark by editing these tokens using another language model. The idea is interesting, and the paper empirically validates the effectiveness of their attack across different watermarks, language models, and datasets. These results clearly establish the effectiveness of the attack in practice." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes an automatic method for editing watermarked text from a language model to evade watermark detection using another (weaker) language model. The paper mainly considers the \"red-green list\" watermark of Kirchenbauer et al. and variants thereof, though the techniques should presumably generalize to other watermarks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The paper distinguishes its main contributions from prior work by arguing that prior work on automatically removing watermarks involved using language models that were at least as strong as the original watermarked language model. However, one notable exception is the work of Zhang et al. [1], who seem to also focus on removing watermarks using weaker language models. This work is cited in the present paper but not discussed in any detail. It would be great if the authors can update their paper with a discussion of how their work differs from [1]. Otherwise, the novelty/significance of the main contributions over prior work is not clear.\n\n\n[1] Zhang et al. (2023) Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models. https://arxiv.org/abs/2311.04378" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- Can the authors provide a baseline that uses the local reference model to do the paraphrase attack?\n- What could be potential adaptive defenses for this attack?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- I find the proposed method very interesting and quite different from the previous work. Meanwhile, the method doesn't require a strong oracle model like a paraphrasing attack, which makes the threat model more realistic.\n- I really enjoy reading this paper, especially section 3.1, which gives readers a lot of insights.\n- The results look positive and a lot of different watermarking schemes are covered (most results are presented in the appendix)." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "In this paper, the authors introduce a novel watermark-removal attack that requires only a small watermark-free reference model. The attacker first estimates the probability of the generated token at position i being in the watermark's green list, which correlates with the relative confidence of the most likely token among the top k tokens. According to the confidence score, the attacker then combines the probability distributions at position i from both the watermarked model and the reference model to sample the token. This approach effectively evades watermark detection while maintaining high text quality." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The proposed method relies on using the logits/output probabilities of the watermarked model. This might limit the attack to some API models that may not return the logits/probabilities or only return top-k probabilities or even calibrated probabilities.\n- The paper uses perplexity or loss to measure the text quality, but I think it's not enough to show the quality of the text. For example, the model can generate an answer for a math question with a very low perplexity, but the answer is completely wrong. So, I think it will be more helpful if the authors can include more text quality metrics like P-SP used in [1] or even a model-based evaluation like asking a large oracle model which generation is preferable.\n- I think it's also helpful to the paper if the answers can show the results under different data distributions instead of overall c4.\n\n[1] Kirchenbauer, J., Geiping, J., Wen, Y., Shu, M., Saifullah, K., Kong, K., Fernando, K., Saha, A., Goldblum, M., & Goldstein, T. (2023). On the Reliability of Watermarks for Large Language Models. ArXiv, abs/2306.04634." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. In Table 1, Watermark (smoothing) has a lower perplexity than Watermark (or even Unwatermark) in some cases (e.g., Llama2-7b). In other words, the attack can even improve the quality of the text, which seems counterintuitive as the reference model is weaker. This also raises a concern about whether perplexity is the right measure to look at the quality of a text here. The authors may want to include other text quality metrics in the numerical studies.\n2. I would like to know if the authors can discuss the potential pitfalls of their methods, e.g., provide concrete examples or scenarios where their smooth attack might fail, and discuss the implications of such failures" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "Many existing methods for statistical watermarking have primarily concentrated on the generation and detection of watermarks. This paper takes a different approach by examining statistical watermarking from a new perspective. This perspective is interesting and may also aid in the development of improved watermark generation and detection techniques." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This work develops a smooth attack in the “green-red list” watermarking framework. The paper shows that a smooth attack makes it easier to bypass the detector while still preserving the quality of the text." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The significance level $S_t$ is unobserved and was estimated using a surrogate quantity, $c_t$. Though the authors showed that there is generally a negative correlation between $c_t$ and $S_t$, this is only a weak justification. It is possible that a small $c_t$ would correspond to a large $S_t$ in some situations, e.g., when $K$ is small. \n2. The method only applies to the “green-red list” watermarking scheme, which is known to be biased because it does not preserve the original text distribution. In contrast, there are unbiased watermarking methods (e.g., Kuditipudi et al., 2023; Aaronson, 2023). It is unclear if the proposed method applies to unbiased watermarking schemes. Perhaps the authors can provide more discussions about how their method might be adapted or extended to work with unbiased watermarking schemes.\n3. The paper lacks a rigorous theoretical analysis of the effect of the smooth attack on the text quality, e.g., bounds on how much the smoothing attack can affect certain text quality metrics." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Why in Figure 1, top-p sampling (right figure) has some points with the total variation distance being 0 or 1, but top-k sampling (middle figure) does not?\n2. How many queries (prefixes) do you use for computing the bin index as described in Lines[261-266]?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The writing is easy to follow.\n\n2. Propose a smoothing attack scheme against statistical watermarking, and show that the significance level $S_t$ is highly correlated with the total variation distance." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a \"smoothing attack\" that bypasses statistical watermarking in large language models (LLMs). By blending outputs from the watermarked model with a weaker reference model, it removes watermarks without impacting text quality on PPL." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The applicability of this method is limited, as obtaining a high-quality reference model is often not possible (e.g. for GPT-4). Additionally, it requires access to token logits, meaning it is not a purely black-box approach as claimed.\n\n2. In line 146. The authors overclaim that their attack is universally applicable to all statistical watermarking schemes. However, many watermarking schemes [1,2,3] do not use a green list, and their proposed method cannot be applied.\n\n3. Additional metrics are needed to better reflect the quality of the generated text. PPL tends to favor a distribution similar to that of the oracle model, which can introduce bias. It would be more informative to include straightforward metrics, such as BLEU in machine translation, to provide a clearer evaluation.\n\n4. The paper lacks key baseline results needed to demonstrate the effectiveness of the proposed method. Naive smoothing using $\\lambda \\tilde{P}(x)+(1-\\lambda) P^{ref}(x)$ can also remove the watermark while preserving part of the text quality.\n\n5. The choice of z-score threshold used in the experiments is unclear. It would be more straightforward to present the true positive rates at specific theoretical false positive rates, providing a clearer understanding of the method’s performance.\n\n6. The experimental settings for certain tests are suboptimal. For instance, in Table 2, the z-score for XSIR and SIR is too low, indicating that the watermark strength in the original watermarked model is insufficient.\n\n[1] Kuditipudi, R., Thickstun, J., Hashimoto, T. and Liang, P., 2023. Robust distortion-free watermarks for language models. arXiv preprint arXiv:2307.15593.\n\n[2] Hu, Z., Chen, L., Wu, X., Wu, Y., Zhang, H. and Huang, H., 2023. Unbiased watermark for large language models. arXiv preprint arXiv:2310.10669.\n\n[3] Dathathri, S., See, A., Ghaisas, S., Huang, P.S., McAdam, R., Welbl, J., Bachani, V., Kaskasoli, A., Stanforth, R., Matejovicova, T. and Hayes, J., 2024. Scalable watermarking for identifying large language model outputs. Nature, 634(8035), pp.818-823." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024watermark,\ntitle={Watermark Smoothing Attacks against Language Models},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1AYrzmDK4V},\nnote={under review}\n}" }, "abstract": { "value": "Statistical watermarking is a technique used to embed a hidden signal in the probability distribution of text generated by large language models (LLMs), enabling the attribution of the text to the originating model. We introduce the smoothing attack and show that existing statistical watermarking methods are not robust against minor modifications of text. In particular, with the help of a weaker language model, an adversary can smooth out the distribution perturbation caused by watermarks. The resulting generated text achieves comparable quality to the original (unwatermarked) model while bypassing the watermark detector. Our attack reveals a fundamental limitation of a wide range of watermarking techniques." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "LLM Watermark" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/c92dc85ed9449585d4fba912de43cd2a2ad0e683.pdf" }, "presentation": null, "primary_area": { "value": "alignment, fairness, safety, privacy, and societal considerations" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/b4c2fe02138d48e982bdf321f62c386d64d3d5a1.zip" }, "title": { "value": "Watermark Smoothing Attacks against Language Models" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "1. Could you clarify what “imputation” refers to in Table 2? Are there results available for CoA without MRFS, and what does “w/ ROUGE” mean? My understanding was that ROUGE is used only in ASQA.\n\n2. In Table 3, could you provide separate statistics for input and output tokens, as well as the average token usage per action? This would help readers better understand the specific cost details.\n\n3. Could you elaborate on what is meant by the term “knowledge boundary”?\n\n4. Are the results of the Chain-of-Action framework directly comparable to previous studies? I noticed that this study used GPT-4, while DSP and SearchChain relied on older-generation LLMs (text-davinci-002 and gpt-3.5-turbo, respectively).\n\n5. Would it be fair and perhaps clearer to rename Sections 2.2.1 and 2.2.2 as \"Data Collection\" and \"Data Verification,\" instead of “Actions Design” and “Actions Workflow”? These alternative terms seem easier to understand and align well with the content of the corresponding subsections." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. This study introduces a framework embodying the divide-and-conquer approach, effectively breaking down complex tasks into manageable components that are tackled sequentially. This structure enhances the model's ability to handle multifaceted queries with improved precision.\n\n2. The empirical results demonstrate notable improvements in both performance and efficiency, as reflected in reduced API calls and token usage compared to prior methods. These gains underscore the framework’s effectiveness and potential for cost-saving in real-world applications.\n\n3. The introduction of the multi-reference faith score (MRFS) is a contribution, which effectively identifies and mitigates information conflicts, and improves answer reliability and trustworthiness in real-time applications." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces the Chain-of-Action (CoA) framework, a novel approach to multimodal and retrieval-augmented question answering that enhances the faithfulness and reasoning quality of large language models (LLMs). CoA addresses key challenges in QA, such as unfaithful responses and weak reasoning, by decomposing questions into a series of reasoning steps or actions that systematically retrieve and verify information from various sources. The framework introduces three \"Plug-and-Play\" actions—web querying, knowledge encoding, and data analyzing—that support multimodal data integration. Additionally, a multi-reference faith score (MRFS) is proposed to resolve inconsistencies and improve response accuracy. Experimental results demonstrate CoA’s effectiveness in handling complex questions across QA benchmarks and in real-world applications, particularly in the Web3 domain." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper’s primary weakness lies in how it presents its key concepts and narrative. Many claims, such as \"multimodal,\" \"plug-and-play,\" and \"action-based\" elements, lack direct evidence or clear definitions, making it challenging to follow the core contributions. Though the pipeline is straightforward, understanding the study's actual workflow is hindered by (1) inaccurate terminology, (2) loosely connected methodology descriptions, and (3) a mix of abstract workflows and technical details.\n\n2. Certain terms are uncommon or seem misapplied, which leads to confusion. For example, terms like \"multimodal\" (when referring to text and tabular data), \"chain-of-action\" (more of a \"chain-of-data-collection-and-verification\"), \"actions design\" (data collection), \"actions workflow\" (data verification), \"node\" (sub-question), and \"knowledge boundary\" (what a model actually knows) lack clarity and could benefit from more precise definitions or alternatives.\n\n3. Question decomposition appears critical to this framework, yet there is limited discussion on decomposition strategies or comparisons with existing baselines. Further elaboration here would strengthen the paper's contributions.\n\n4. The \"plug-and-play\" feature is presented as a low-cost prompting strategy; however, integrating retrieval for each data type (e.g., web, internal knowledge) may not be straightforward. It may be worth reconsidering or refining this claim to better reflect its implementation complexity.\n\n5. The paper’s claim of multimodal data handling is unclear. If the input consists of real-time information, domain knowledge, and tabular data, it may be more accurately described as handling heterogeneous data rather than multimodal data. Additionally, if tabular data is linearized as text for LLM input, the fundamental multimodal claim weakens.\n\n6. The study does not include ablations to show the specific contribution of tabular data. Providing such analyses could clarify its impact on the framework's performance.\n\n7. Section 3.2 mentions expert evaluation on a 1 to 3 scale based on three criteria, but it lacks details on the expert recruitment process, qualifications, and any inter-rater reliability metrics. Adding these details would increase the transparency and credibility of the evaluation process." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Can you elaborate on the key differences between \"thoughts\" in CoT and \"actions\" in CoA? How does this change improve the overall performance? It would also be helpful if you can discuss the limitations and trade-offs between them.\n2. If the system doesn't have the ability to add additional actions like web query, does CoA still perform better than CoT. \n3. Does CoA add significant latency to QA process?" }, "rating": { "value": 8 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The authors utilize newer multimodal LLM abilities to perform actions such as web query and data analysis. The authors come up with a new QA mechanism for LLMs which uses the actions The method is called Chain of Action(CoA).\n2. The authors demonstrate that this method significantly outperforms other reasoning and QA methods on many QA datasets. \n3. The improvement of using actions over thoughts does seem to be the natural way of solving a question. This approach has significant potential for improving QA capabilities of LLMs." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors propose a new QA retrieval mechanism called Chain of Action(CoA). When a question is asked to an LLM, there is a prompt which generates a list of actions the LLM needs to take first to effectively answer the questions. They introduce a Plug and Play approach where in case of Multimodal LLMs, the actions taken can be integrated into the application. The actions can be web query or data analysis. The paper integrates 3 such actions. The LLM then performs each of the individual action generated and then there is another query which combines information from all the actions. The LLM then gives an answer based on the newly injected information" }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Based on the number of actions to be taken and what kind of \"plug\" is used for the action, the time taken to finish all actions and send out an answer might become significant. It would have been good to see the study on latency(eg. average response time) of the system because of the new method.\n2. It would be helpful to conduct an ablation study when you remove specific action types to isolate their impact on performance. This would provide clearer insights on how much this method relies on additional capabilities. \n3. Comparing CoA with and without the ability to perform additional \"plugs\" across different types of questions can be useful in understanding the impact of this method." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "1. Can the authors provide more details on how the CoA framework could be adapted for tasks involving visual or mixed data modalities?\n2. How does the framework handle discrepancies or conflicts when sources provide contradictory information?\n3. Are there plans to explore CoA's performance in real-time, fast-evolving information retrieval scenarios where data may change rapidly (e.g., live news events)?\n4. Could the use of CoA extend to tasks requiring intricate reasoning paths that involve recursive or nested logic?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- **Innovative Framework**: The CoA's structured decomposition into sub-questions and its use of domain-adaptable plug-and-play actions represent a significant advancement in enhancing the faithfulness and accuracy of LLM responses.\n- **Empirical Validation**: Demonstrated strong performance on benchmarks and real-world applications, notably outperforming existing baselines in multimodal QA tasks.\n- **Verification Mechanism**: The multi-reference faith score is an effective metric for cross-validating LLM-generated answers against external sources, enhancing reliability.\n- **Practical Impact**: Real-world implementation in a Web3 QA system showed increased user engagement and positive feedback, validating the method's applicability." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents the Chain-of-Action (CoA) framework designed to improve large language models' (LLMs) performance in multimodal and retrieval-augmented question-answering (QA). CoA addresses challenges such as hallucination and weak compositional reasoning by decomposing questions into reasoning chains. This method incorporates plug-and-play actions for retrieving diverse data sources and uses a novel multi-reference faith score for verification. Empirical results show CoA outperforms other methods in public benchmarks and real-world applications." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- While the CoA approach shows strong empirical performance, its adaptability to more diverse or unstructured data modalities beyond text and tabular data remains to be proven.\n- The scalability and efficiency when integrating more complex or real-time data sources require further exploration, especially in scenarios with rapidly changing information.\n- The approach, despite its modular design, may face challenges in tasks involving higher-order reasoning or complex multi-step dependencies that are not purely fact-based." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024chainofaction,\ntitle={Chain-of-Action: Faithful and Multimodal Question Answering through Large Language Models},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1BdPHbuimc},\nnote={under review}\n}" }, "abstract": { "value": "We present a Chain-of-Action (CoA) framework for multimodal and retrieval-augmented Question-Answering (QA). Compared to the literature, CoA overcomes two major challenges of current QA applications: (i) unfaithful hallucination that is inconsistent with real-time or domain facts and (ii) weak reasoning performance over compositional information. Our key contribution is a novel reasoning-retrieval mechanism that decomposes a complex question into a reasoning chain via systematic prompting and pre-designed actions. Methodologically, we propose three types of domain-adaptable `Plug-and-Play' actions for retrieving real-time information from heterogeneous sources. We also propose a multi-reference faith score to verify conflicts in the answers.\nIn addition, our system demonstrates that detecting the knowledge boundaries of LLMs can significantly reduce both LLM interaction frequency and tokens usage in QA tasks. Empirically, we exploit both public benchmarks and a Web3 case study to demonstrate the capability of CoA over other methods." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "large language model", "question answering", "chain-of-thought" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/e4915d4d96a9ccd999db7a92c8aef32d5ae71c8f.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Chain-of-Action: Faithful and Multimodal Question Answering through Large Language Models" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": { "value": "I have read and agree with the venue's withdrawal policy on behalf of myself and my co-authors." } }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Is the effectiveness of IOM and DOA useful in other tasks such as multiple object tracking?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Application-oriented Innovation: O2VIS introduces occupancy information into the memory update process, making it possible to maintain object identity consistency more accurately in scenes where objects frequently appear and disappear. This occupancy-aware memory management strategy provides a useful enhancement for video instance segmentation.\n\n2. Empirical Support: The experimental results on various benchmark datasets show improved average precision (AP) and average recall (AR), supporting the method's effectiveness. Additionally, the ablation studies validate the contributions of IOM and DOA, strengthening the reliability of the results.\n\n3. Well-designed Component Structure: The decoupled approach in DOA separately manages existing and new objects, using occupancy-guided Hungarian matching to reduce incorrect associations. This is a practical and effective design choice." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents the O2VIS framework, aiming to improve long-term consistency in video instance segmentation. By introducing Instance Occupancy Memory (IOM) and Decoupled Object Association (DOA), this method enhances the stability of object tracking in dynamic video scenes, effectively differentiating between recurring and new objects. The paper demonstrates the approach's performance on multiple benchmark datasets, such as YouTube-VIS and OVIS, highlighting its advantages in accuracy." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Limited Theoretical Innovation: The calculation of foreground probability is essentially a weighted adjustment using output probabilities and does not introduce a novel computational framework or algorithm. IOM and DOA represent more of an applied enhancement of existing memory and association techniques, rather than a fundamental theoretical breakthrough, which may limit the impact at conferences focused on theoretical innovation.\n\n2. Unclear Generalizability: The method is primarily designed for video instance segmentation, and its applicability to other tasks, such as multi-object tracking, has not been demonstrated. Verifying IOM and DOA’s effectiveness in other visual tasks would strengthen the paper’s generalizability.\n\n3. Dependence on Pre-trained Model Accuracy: Since the foreground probability relies on classification outputs, errors in these outputs could lead to incorrect memory updates, potentially destabilizing tracking performance. This dependency might reduce overall system stability, particularly when applied to longer or more complex video sequences." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 1 }, "primary_area": null, "questions": { "value": "- It is very hard to understand Figure 1. There are barely any labels and barely any text in the caption to explain what each of the icons in the figure means. The first teaser figure should be very easy to understand and should convey an overall takeaway from the method, and not describe the method itself. \n- Can you explain how you get an object's occupancy O near L206?\n- If an object's occupancy is 0, why should it's new feature representation be added to the memory?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The motivation is well-grounded that the objects should be treated differently based on if they have been seen before" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents a method for long-term tracking consistency in videos for the task of video instance segmentation. The core idea is that using the visibility or occupancy of the objects can help in associating their features correctly so as to differentiate between new and previously seen objects. Treating these two kinds of objects differently by associating them separately also helps. Experiments exist that compare state-of-the-art approaches to the proposed method." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- One of the most important baselines is missing for this task is SAM2 for video instance segmentation\n- While the motivation is good, similar ideas have appeared before for tracking, for instance in DeepSORT or Detecting Invisible People. These are not segmentation approaches but should be cited to acknowledge that both of the main contributions of this paper have appeared before.\n- It seems like the ID switches metric from multi-object tracking based on bounding boxes, is what the paper wanted to improve but there is no comparison to prior approaches with that metric so it is hard to tell if their claim of long-term consistency is valid over an entire dataset." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 2 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "see Weaknesses" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "Innovation: The study proposes an instance occupancy memory mechanism that addresses challenges in maintaining consistency when objects disappear and reappear, making it well-suited for complex, dynamic video scenes.\nPerformance: The experimental results show that O2VIS significantly outperforms current state-of-the-art methods across multiple datasets, especially in AP scores.\nDecoupled Strategy: By implementing a decoupled association strategy for handling existing and new objects separately, the method avoids common background misalignment issues, enhancing tracking accuracy.\nComprehensive Experiments: The paper provides thorough experimental comparisons with existing VIS methods, and ablation studies validate the effectiveness of each technical component, demonstrating the contributions of IOM and DOA modules." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces O2VIS, a novel framework for video instance segmentation that enhances long-term consistency in object tracking. The framework incorporates an Instance Occupancy Memory (IOM) module and a Decoupled Object Association (DOA) strategy, effectively distinguishing between new and recurring objects across frames. By decoupling the association of existing and newly appeared objects, the method maintains stable and consistent object identities throughout videos. Experimental results demonstrate that O2VIS achieves state-of-the-art AP scores on the YouTube-VIS 2019, 2021, and 2022 datasets, setting a new benchmark for the VIS task." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "Insufficient Details: While the paper introduces the Occupancy-guided Hungarian Matching and Decoupled Object Association strategies, implementation details are limited. Providing pseudocode or more concrete algorithmic descriptions could enhance clarity.\nComputational Cost: The addition of IOM and DOA likely increases computational complexity, particularly due to multi-frame memory updates and association matching. It would be beneficial to quantify the computational overhead of these modules within the paper.\nGeneralizability: Experiments are currently focused on standard datasets like YouTube-VIS. The model’s performance in more challenging scenarios, such as high occlusion or rapid object movement, remains unclear.\nModel Complexity: With the integration of multiple modules, the overall model structure is complex, which may pose deployment challenges. Future work could explore simplifying the model or improving its efficiency." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1 The authors should include a comprehensive comparison of computational resources, including model parameters, inference speed, and memory usage, with existing methods. This would provide crucial context for understanding the practical trade-offs of their approach. \n\n2 Additionally, including more detailed pseudo-code for key algorithms and visualizations of memory usage patterns would enhance the technical clarity of the paper. \n\n3 Finally, an analysis of failure cases and performance on longer video sequences would provide valuable insights into the method's limitations and potential areas for improvement." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1 The paper's technical contributions are both novel and well-executed. The IOM mechanism provides an elegant solution to the challenging problem of maintaining object identity consistency, while the decoupled association strategy effectively addresses the issue of new object appearances. \n\n2 The comprehensive experimental evaluation, including extensive ablation studies, convincingly demonstrates the effectiveness of each component. The strong performance across multiple benchmarks further validates the proposed approach." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces O2VIS, a novel framework designed to enhance long-term consistency in video instance segmentation. The work presents two main technical innovations: an Instance Occupancy Memory (IOM) for tracking global instance features and their occupancy status, and a Decoupled Object Association (DOA) strategy that separately handles existing and new objects. The framework demonstrates state-of-the-art performance on YouTube-VIS benchmarks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1 The paper does not provide comparisons of model parameters and inference speed with existing methods, making it difficult to assess the practical implications of implementing this approach. \n\n2 There is no discussion of memory consumption or runtime benchmarks, which are crucial considerations for real-world applications. \n\n3 Some technical details, particularly regarding the IOM update mechanism and the interaction between TE and TA trackers, could be explained more thoroughly." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "Please see Weaknesses." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The paper is well-written and easy to follow. A thorough experimental analysis has been performed." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces an occupancy memory and a decoupled object association to track global features of objects long term and to ensure consistent matching of new and old objects. The proposed method achieves good performance on the VIS task." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Limited technical novelty: The paper proposes some techniques to improve VIS, but all of these techniques have been seen in other tracking/VIS works in some form or the other. For example, the IOM is similar to global track queries in trackformer [2], Hungarian matching to align current objects to a global memory in DOA has been explored before in many works.\n2. Incremental improvement: The results in table 1 and 2 often show a minimal improvement as compared to prior works. For example, on the youtube vis 2019 dataset, the method only gets a 0.2 points improvement over DVIS++ using the R50 backbone. Similar trend is observed for other datasets and other backbones. These improvements could often just come from randomness during training, so it would be nice if the authors could put error bars in the tables to demonstrate consistency. \n3. Some prior works (e.g., CAROQ [1]) use query-based propagation for global tracking. How does the proposed method compare with such a method in terms of the number of parameters involved in tracking and the tracking speed? The proposed method requires 2 networks for tracking, as opposed to 1 network in most prior works, so some comparison table on the average time taken and the parameters involved solely for tracking would also be insightful.\n4. There are some typos in the paper, e.g., a capitalized letter mid-sentence in line 47.\n\n\n[1] Choudhuri et al., Context-Aware Relative Object Queries to Unify Video Instance and Panoptic Segmentation, CVPR 2023\n[2] TrackFormer: Multi-Object Tracking with Transformers, Meinhardt et al., CVPR 2022" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We propose an occupancy-guided memory and temporally consistent object association pipeline." }, "_bibtex": { "value": "@misc{\nlee2024textotextvis,\ntitle={\\${\\textbackslash}text\\{O\\}\\_{\\textbackslash}text\\{2\\}\\${VIS}: Occupancy-aware Object Association for Temporally Consistent Video Instance Segmentation},\nauthor={Seunghun Lee and Jiwan Seo and Minwoo Choi and Kiljoon Han and Jaehoon Jeong and Ehsan Adeli and Sang Hyun Park and Sunghoon Im},\nyear={2024},\nurl={https://openreview.net/forum?id=1BlEVFmqwn}\n}" }, "abstract": { "value": "In this paper, we present Occupancy-aware Object Association for Video Instance Segmentation ($\\text{O}_{\\text{2}}$VIS), a new framework crafted to improve long-term consistency in instance tracking. We introduce the Instance Occupancy Memory (IOM) that tracks global instance features and their occupancy status to effectively differentiate between recurring and new objects. It ensures consistent tracking and effective management of object identities across frames, enhancing the overall performance and reliability of the VIS process. Moreover, we propose a Decoupled Object Association (DOA) strategy that handles existing and newly appeared objects separately to optimally assign indices based on occupancy. This technique enhances the accuracy of object matching and ensures stable and consistent object alignment across frames, especially useful in dynamic settings where objects frequently appear and disappear. Extensive testing and an ablation study confirm the superiority of our method over traditional methods, establishing new standards in the VIS domain." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": { "value": [ "~Seunghun_Lee3", "~Jiwan_Seo2", "~Minwoo_Choi1", "~Kiljoon_Han1", "~Jaehoon_Jeong1", "~Ehsan_Adeli1", "~Sang_Hyun_Park1", "~Sunghoon_Im1" ] }, "authors": { "value": [ "Seunghun Lee", "Jiwan Seo", "Minwoo Choi", "Kiljoon Han", "Jaehoon Jeong", "Ehsan Adeli", "Sang Hyun Park", "Sunghoon Im" ] }, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Video instance segmentation", "Long-term memory", "Temprorally consistent learning" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": { "value": "lee|\\texto_\\text2vis_occupancyaware_object_association_for_temporally_consistent_video_instance_segmentation" }, "pdf": { "value": "/pdf/2559f6b728d197bf558232ea5fc8abb942b31cc9.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "$\\text{O}_\\text{2}$VIS: Occupancy-aware Object Association for Temporally Consistent Video Instance Segmentation" }, "venue": { "value": "ICLR 2025 Conference Withdrawn Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Withdrawn_Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Why in Figure 1 and Figure 2, the same image has two different retrieved CAD models?\n2. Can you provide the results of the error based on the quality of the retrieved CAD model?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The task of predicting the 6D pose of internet videos without additional prior is important for a lot of downstream tasks.\n2. The whole pipeline is reasonable, fetch the similar CAD model and do rough alignment. Then further leverage the 2D tracking results to get the smoothed trajectories, that are more motion-consistent across time.\n3. The experiments on the retargeted motion on robotics further show the usefulness of the extracted smoothed trajectories." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a pipeline for extracting the 6D pose trajectory from an internet video without the need of the CAD for the specific object. The authors leverage vision features to retrieve the most similar CAD model of the object, then do per-frame alignment leveraging the same vision features of the original image and rendered from the CAD. They further estimate the rough object size using LLM and leverage 2D tracking models to get inter-frame rotation consistency. The authors conduct experiments and demonstrate their superior performance. They also show demos that their trajectory can be retargeted to guide the movement of the robot." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The authors demonstrate that compared to model-based methods, whose performances suffer from the inaccurate CAD mode, their method addresses the challenge. However, there is lack of experiments compared to SOTA model-based methods with their fetched CAD models (e.g. FoundationPose with their retrieved CAD model).\n2. In the 6D pose alignment part, the method applies a sapling-based trajectory to get the rotation, which potentially limits the accuracy of the rotation. In the results figure, there are some rotation errors, not sure if due to the sampling-based strategy or the DINO feature extractor.\n3. For the robotics demo, the end-effector position control is on 6D pose or only on the rotation? From the Figure 9, the translation of the end-effector seems not consistent with the original video and in the simulator" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- l. 323, are the ground-truth meshes contained in the object datasets? \n- Table 1, was the same scale estimate for the meshes used for MegaPose and GigaPose like for the proposed method? \n- Which dynamics model is used for the optimization problem in eq 4? How is tracking of the optimized trajectory implemented?\n- See additional questions in sec. \"Weaknesses\"." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- The proposed approach for detecting and estimating 6D motion of unknown objects from RGB images is novel and interesting.\n- The paper is well written and easy to follow.\n- The set of experiments demonstrate the shape retrieval and pose estimation well and also compare with state of the art methods.\n- A qualitative example is provided with a real robot which show the robot pouring from one object to another." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a new approach to detect and track the 6-DoF pose of unknown objects from RGB video. The approach is motivated by robot imitation learning from internet video. The approach uses off-the-shelf open-set object detectors, foundation models for segmentation, vision-language (CLIP), and visual features (DINOv2) to detect objects, retrieve similar shapes from a database of CAD models, and matching the object image with a set of rendered views of the object CAD model to estimate 3D orientation. Experimental evaluation is performed quantititvely on YCB-Video and HOPE-Video datasets and a comparison is made with state of the art object detectors for unseen objects for which the CAD model is assumed known (MegaPose, GigaPose). Also, qualitative results on EPIC-Kitchen, and an example of executing the estimated object trajectories on a real robot are shown." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- l. 197ff, CAD model retrieval by rendering views and calculating visual features seems expensive in both, the database generation and the retrieval stage for large datasets such as Objaverse-LVIS. What is the retrieval time for these datasets and how is it implemented to make retrieval efficient?\n- l. 220ff proposes to retrieve rotation by matching to a set of rendered views. What is the choice of N in the experiments? What is the avg/std angular distance between sampled rotations?\n- l. 243ff, the way to prompt the LLM in the supplementary is an offline procedure to collect size estimates for approximately 2200 objects. In the main paper, the description reads as if the LLM is prompted for each detected object using the CLIP text classification. Please describe this more clearly. What if the detected object is not included in the offline calculated set ? \n- l. 286, was estimating the motion of the camera relative to the static background evaluated in this work ? Please clarify.\n- The optimization problem in eq 4 does not provide a description of the used system dynamics model. \n- l. 361, please write more clearly, that while a similar mesh is known, the retrieved mesh does not exactly correspond to the ground truth mesh which is an assumption used for MegaPose and GigaPose. \n- Please introduce the pCH metric formally, at least in the supplemental material. The current description is insufficient.\n- l. 519ff, the real robot experiment is rather anecdotal and lacks important details in its descriptions and quantitative evaluation (e.g., success rate). How are the observed object trajectories transfered to the real robot experiment incl. considering the change of view point and embodiment? How does the robot know where the manipulated objects are and how is this matched to the observed object motion? \n- Fig. 8, in the upper additional qualitative result, the bowl object pose is not correctly tracked. Why does the robot still turn the object in a quite different angle ?\n\nAdditional minor comments:\n- Fig. 6, rightmost real robot image seems to be a repetition of the image next to it. Was the wrong image included?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. It might be great if the authors could ablate on the performance variation under different LLMs. Currently it only applies GPT-4, but it is important to know how different LLMs might influence the performance (i.e. one GPT-3.5 & one open-source LLM).\n2. What's the efficiency & cost of such pipeline when performing inference on a 1-minute Instructional videos? \n3. Using a CAD model can be costly since it requires a large database to store predefined meshes, and in open-world scenarios, finding an exact match is often unlikely. However, numerous approaches avoid relying on CAD models. For instance, \"6DGS: 6D Pose Estimation from a Single Image and a 3D Gaussian Splatting Model\" [ECCV 2024]. Have you tried experimenting with such methods? Or say, how do you envision those methods' strengths and weaknesses compared to your method.\n4. For the standard evaluation, it might be beneficial to add another dataset evaluation using different cameras, say iPhone sensor as proposed in \"Robust 6DoF Pose Estimation Against Depth Noise and a Comprehensive Evaluation on a Mobile Dataset\" to further validate the approach's generalizability." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "The impact of the paper is dominant in the way that it provides an envision of enriched data for robotic manipulation without human labor force to construct the specific datasets. The methodology is intuitive and the performance enhancement is non-trivial. The paper is overall well-written." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors present a novel approach to extract temporally consistent 6D pose trajectories of manipulated objects from Internet videos to be applied with robotic manipulation task. It tackles the challenges posed by uncontrolled capture conditions, unknown object meshes, and complex object motions. Their evaluation on YCB-V and HOPE-Video datasets shows state-of-the-art performance, with successful motion transfer to a robotic manipulator in both simulated and real-world settings." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "My primary concern lies with the methodological novelty, as the approach largely involves applying an existing pipeline to internet videos. Specifically, the use of an LLM for estimating object scale may be questionable, given potential uncertainties around its accuracy in providing a realistic scale for each object. Aside from this, the methodology essentially adapts previous methods to fit the proposed pipeline. Given these factors, I feel this work might not align with ICLR's focus but could be more suited to a robotics conference." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1 With the similar CAD model retrieval, the classification can also be obtained. I wonder if it is possible to use the CAD model to perform classification directly?" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1 The pose estimation method by retrieving a CAD model, aligning the retrieved CAD model with the object, and grounding the object scale with respect to the scene.\n\n2 Consistent 6D pose trajectory estimation from Internet videos and retargeting trajectories to a robotic manipulator.\n\n3 The pose estimation improvement on YCB-V and HOPEVideo datasets, and transfer from 6D object motion to a 7-axis robotic manipulator." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper pays attention on 6D pose trajectory estimation of a manipulated object from an Internet instructional video with a novel framework. The framework first predicts the 6D pose of any object by CAD model retrieval. Then the smooth 6D object trajectories are extracted and retargeted via trajectory optimization into a robotic manipulator. Experiments on YCB-V and HOPE-Video datasets demonstrate the improvements over RGB 6D pose methods. Moreover, the 6D object motion can be transferred to a 7-axis robotic manipulator." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1 The original contributions should be expressed more clearly. In the proposed method, various existing methods are employed. It is suggested to clearly distinguish the original contributions in this paper and usage of other methods. Specifically, the first contribution locates in the pose estimation method by retrieving a CAD model, aligning the retrieved CAD model, and grounding the object scale with respect to the scene. The subsequent question is that what is the original contribution, the whole pipeline or the detailed design of a particular module? The authors are suggested to express this more clearly in the revised version. For the second and third contributions, it is also recommended to present more clear expressions. \n\n2 For robotic manipulation, the running time of the pose estimation method is a key factor. The proposed method in the paper is somewhat time-consuming with 2s for detector, retrieval and scale estimation per scene and 0.2s for pose estimation per object. To further improve the paper, two suggestions are given. For one thing, the comparaions with other methods on running time are suggested to add. For another, more analysis about the running time is also preferred, such as the recommendations for accelerate the whole method." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "A method to estimate 6D pose and trajectory of an object in the Internet video" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024d,\ntitle={6D Object Pose Tracking in Internet Videos for Robotic Manipulation},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1CIUkpoata},\nnote={under review}\n}" }, "abstract": { "value": "We seek to extract a temporally consistent 6D pose trajectory of a manipulated object from an Internet instructional video. This is a challenging set-up for current 6D pose estimation methods due to uncontrolled capturing conditions, fine-grained dynamic object motions, and the fact that the exact mesh of the manipulated object is not known. To address these challenges, we present the following contributions. First, we develop a new method that estimates the 6D pose of any object in the input image without prior knowledge of the object itself. The method proceeds by (i) retrieving a CAD model similar to the depicted object from a large-scale model database, (ii) 6D aligning the retrieved CAD model with the input image, and (iii) grounding the absolute scale of the object with respect to the scene. Second, we extract smooth 6D object trajectories from Internet videos by carefully tracking the detected objects across video frames. The extracted object trajectories are then retargeted via trajectory optimization into the configuration space of a robotic manipulator. Third, we thoroughly evaluate and ablate our 6D pose estimation method on YCB-V and HOPE-Video datasets and demonstrate significant improvements over existing state-of-the-art RGB 6D pose estimation methods. Finally, we show that the 6D object motion estimated from Internet videos can be transferred to a 7-axis robotic manipulator both in a virtual simulator as well as in the real world. Additionally, we successfully apply our method to egocentric videos taken from the EPIC-KITCHENS dataset, demonstrating potential for Embodied AI applications." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "6DoF pose estimation", "robotic manipulation from video" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/615dd70e5b1540a6e035285198f7dd6fdd5c5d76.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/25538edbe05f4d6d9b33e61503a69a7887b0fc84.zip" }, "title": { "value": "6D Object Pose Tracking in Internet Videos for Robotic Manipulation" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "Q1. The proposed architecture adds significant computational cost to the internal representation of the model compared to vanilla transformers and some of the previously proposed models. It seems this does not have a significant effect on the training time compute and memory complexity of the model. Have the authors conducted any studies to compare the inference-time cost of TM++ compared to other methods? \n\nQ2. As mentioned by authors, some time series tasks (imputation, anomaly detection) benefit more from diverse representations while others like forecasting and classification benefit from consistent representation. Given this, is there any way to leverage a routing model dependent on the proposed task type, which could lower the inference-time cost of this model?\n\nQ3. MTS-Mixer (Li et. al., 2023) presents another approach to channel decomposition which similarly outperformed competing models, but they found the approach worked best with MLPs rather than attention-based models. Have the authors explored this technique for separating from attention mechanisms which could lead to further efficiency and model performance?" }, "rating": { "value": 10 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "S1. The proposed model captures both short- and long-term dependencies by transforming time series data into multi-resolution images, enabling the analysis of complex temporal and frequency-domain patterns that challenge traditional models. The authors validate this with experimental results showing the new architecture outperforms SOTA models on most standard benchmarks. The ablation study helps validate the importance of the individual parts of the architecture – the channel mixing, image decomposition, and multi-scale and multi-resolution mixing. This approach continues to validate the benefits of integrating image analysis techniques with time series tasks.\n\nS2. The architecture is flexible for supporting different kinds of time-series tasks. The hierarchical multi-scale and multi-resolution mixing modules enable the model to flexibly adapt across various time series tasks, from forecasting to anomaly detection, promoting robust and accurate performance across applications.\n\nS3. Empirical Validation: the testing in this paper on eight benchmark time series tasks, including the hyperparameter ablation results, shows TIMEMIXER++ consistently surpasses both general-purpose and task-specific models, affirming its potential as a high-performance, general-purpose solution for time series analysis. The experiments were very thorough." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents TIMEMIXER++, a general-purpose model for various time series tasks, including forecasting, classification, anomaly detection, and imputation. Utilizing a multi-scale, multi-resolution framework, the proposed method transforms time series data into multi-resolution images to capture complex temporal and frequency-domain patterns, enabling flexibility across analytical applications. The model’s approach includes dual-axis attention for decomposing seasonal and trend components and hierarchical multi-scale and multi-resolution mixing to integrate patterns across scales. The proposal achieves strong performance across eight benchmark tasks, outperforming both general-purpose and task-specific models. This work contributes to advancing time series analysis with new state-of-the-art benchmarks across settings." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "W1. There is little exploration of scaling of model size, which would be an interesting avenue for validating the model architecture in a zero shot setting. The current zero-shot experiments are primarily in-domain and not cross-task." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 4 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See weaknesses (W2, W3, W4)." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. the authors introduce a robust framework TimeMixer++ that leverages multi-resolution time imaging, multi-scale mixing, and dual-axis attention to enhance general time series analysis. They present SOTA results on four different tasks.\n2. the integration of both multi-scale and multi-resolution mixing strategies for adaptive pattern extraction demonstrates innovation.\n3. the manuscript and appendix are well-prepared, but the authors have not yet released the promised code." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents TimeMixer++, an advanced framework designed to enhance general time series analysis. TimeMixer++ integrates multi-resolution time imaging, multi-scale mixing, and dual-axis attention mechanisms to effectively capture and adapt to diverse patterns within time series data. This innovative approach allows for robust and flexible analysis across various temporal scales and resolutions." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. the fonts in the figures should be enlarged for better readability. For example, in Figure 1 (right), the label \"Benchmarking model performance across representation analysis in four tasks\" appears blurred. Additionally, consider using a single set of legends for all four tasks to enhance clarity.\n2. the source code repository has not released for reproducing, i will consider raising the score if the released repository and the consistency of the results.\n3. more detail on how it compares to recent models like TimesNet and iTransformer on specific time series tasks would strengthen the paper’s claims.\n4. including a discussion on computational efficiency (e.g., FLOPs, memory usage) for different tasks could enhance the paper’s utility." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "[1] Questions follow from the points listed in the weakness section.\n[2] What is the naming of the method, i.e., TimeMixer++ in terms of? or just because both the methods target processing multi-scale time series? \n[3] What is the connection with the method TimeMixer?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "The methods used in the paper (e.g., time imaging and image decomposition) are very interesting. The evaluation is comprehensive: the authors discuss long and short-term forecasting, zero-short forecasting, classification, and anomaly detection." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a time series pattern machine method called TimeMixer++ for processing multiscale time series. The method transforms time series into multi-resolution time images to enable pattern extraction with respect to temporal and frequency domains followed by three dominant modules - (1) input projection; (2) a stack of Mixerblocks; (3) output projection. Extensive experiments show the proposed method obtains improvement over the well-established competing methods." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "In terms of the forecasting results shown in Tables 3 and 4, the performance gain is negligible, and such minor improved performance certainly can be attributed to the parameter tuning, e.g., a well-tuned parameter settings for TimeMixer++ while a weak parameter settings for other competing methods. \n\nThe paper barely offers insights both theoretically and experimentally. The theoretical understanding of the improvement as well as its time imaging and multi-resolution mixing is lacking, mostly based on intuition and simply blending the models. \n\nThere are some papers that already discussed the use of frequency analysis and the frequency components extraction for model deployment (e.g., [1][2][3]) to capture the periodic patterns, and they all claim it can capture the global interaction and patterns among time series, so what is the benefits of introducing multi-resolution time imaging, and it is worthwhile to compare them in ablation study? In addition, it is encouraged to cite the papers [1][2][3] if not yet in the references. \n\n\n\nReferences:\n[1] Koopa: Learning Non-stationary Time Series Dynamics with Koopman Predictors https://arxiv.org/pdf/2305.18803\n[2] TFDNet: Time-Frequency Enhanced Decomposed Network for Long-term Time Series Forecasting https://arxiv.org/abs/2308.13386\n[3] FEDNET: FREQUENCY ENHANCED DECOMPOSED NETWORK FOR OUT-OF-DISTRIBUTION TIME SERIES CLASSIFICATION https://openreview.net/forum?id=OVu9DsOjgH" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "TimeMixer++ is a time series pattern machine that employs multi-scale and multi-resolution pattern extraction to deliver SOTA performance across 8 diverse analytical tasks, including forecasting, classification, anomaly detection, and imputation." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024timemixer,\ntitle={TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1CLzLXSFNn},\nnote={under review}\n}" }, "abstract": { "value": "Time series analysis plays a critical role in numerous applications, supporting tasks such as forecasting, classification, anomaly detection, and imputation. In this work, we present the time series pattern machine (TSPM), a model designed to excel in a broad range of time series tasks through powerful representation and pattern extraction capabilities. Traditional time series models often struggle to capture universal patterns, limiting their effectiveness across diverse tasks. To address this, we define multiple scales in the time domain and various resolutions in the frequency domain, employing various mixing strategies to extract intricate, task-adaptive time series patterns. Specifically, we introduce \\method, a general-purpose TSPM that processes multi-scale time series using (1) multi-resolution time imaging (MRTI), (2) time image decomposition (TID), (3) multi-scale mixing (MCM), and (4) multi-resolution mixing (MRM) to extract comprehensive temporal patterns. MRTI transforms multi-scale time series into multi-resolution time images, capturing patterns across both temporal and frequency domains. TID leverages dual-axis attention to extract seasonal and trend patterns, while MCM hierarchically aggregates these patterns across scales. MRM adaptively integrates all representations across resolutions. TimeMixer++ achieves state-of-the-art performance across 8 time series analytical tasks, consistently surpassing both general-purpose and task-specific models. Our work marks a promising step toward the next generation of TSPMs, paving the way for further advancements in time series analysis." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "time series", "pattern machine", "predictive analysis" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/f2234161056e92e06729fbbd68ea8ecfb4ae47f8.pdf" }, "presentation": null, "primary_area": { "value": "learning on time series and dynamical systems" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "* How does the method presented in the paper differs from existing works?\n* Figure 3, Table 1, Figure 4: A baseline where the parameters are randomly selected is needed. What are the performances of such a baseline?\n* Table 1, Figure 3: How does the method compares to other pruning methods such as the one presented in this survey [2] \n* Table 2: How does the method compares to other model merging method such as the one presented in this survey [3] \n\n\n[2] https://arxiv.org/pdf/2308.06767\n[3] https://arxiv.org/pdf/2408.07666" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper presents interesting and novel applications of saliency maps to model quantization and model merging.\n\nThe paper strongly motivates the usage of saliency maps." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "A white-box method for identifying the most important/salient regions of an input for making a prediction is presented. The paper describes the method using the formalism of differential geometry and apply it to VLMs and LLMs. For VLMs, they apply their method to sensitivity analysis For LLMs, they apply their method to model quantization and model merging." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "I currently suggest to reject this paper on the basis that I don't understand the novelty of the method and due to the lack of comparisons with the baselines. I am open to changing my mind, but I strongly encourage the authors to focus on those specific points in their response and be very clear how their method differs from existing works.\n\nNovelty of the method: Not clear how the proposed method is different from other analysis methods such as Saliancy maps [1] and adversarial perturbations. A throughout analysis of the related methods should be presented and compared against in the paper.\n\nComparison with the baselines: Several compelling applications of the method are proposed, but no comparison to existing baseline methods tacking these applications. The authors should consider comparing the presented method against relevant baselines on standard benchmarks so that the reader can assess the usefulness of the method.\n\nClarity of the paper: I found the paper to be hard to follow. The paper introduces unnecessarily abstracts notions to describe the method. I don't understand why such abstraction is needed to describe the idea presented in the paper. Moreover, a lot of terms a unnecessarily defined. For example, $l(\\omega|y,x,theta)$ could be written as $\\log P(y|x,\\theta,\\omega)$ and $f(\\omega)$ as $-P(y_\\text{pred}|x,\\theta,\\omega$ and it would make the reading clearer. Some terms like $h_j$ are not clearly defined.\n\n[1] https://arxiv.org/abs/1312.6034" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "See Weaknesses" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "+ FI offers a theoretically grounded approach to assessing parameter and input sensitivity, which can support robustness improvements across applications.\n\n + The paper provides experiments on various models, including applications of FI in quantization and model merging, demonstrating practical value." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a \"First-order Local Influence\" (FI) metric for quantifying LLM and VLM sensitivity to perturbations. By examining both internal (parameter) and external (input) perturbations, the FI metric aims to identify model weaknesses and improve robustness through selective parameter preservation. Experiments demonstrate FI’s potential in tasks like model quantization and merging." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The paper is positioned primarily around LLMs and VLMs, but these stability concerns are more broadly applicable to general ML. A broader contextual framing would benefit the paper.\n\n- The choice to protect high-FI parameters during distillation/model merging is questionable since some high-FI parameters might correspond to irrelevant or “nonsensical” inputs.\n\n- Prior works on Fisher Information Matrix (FIM) in pruning and parameter sensitivity (e.g., Frantar & Alistarh, 2023; Yu et al., 2024) and Sharpness-Aware Minimization (SAM) (Foret et al., 2021) are not mentioned. These are relevant for contextualizing FI's robustness contributions." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See weaknesses.\n\n- Can you provide more details about the compute cost of your approach?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- The proposed approach seems effective in identifying input regions/parameters that have a large effect on model output. \n- The authors test their approach in a range of applications. \n- I like the application of identifying sensitive parameters that should remain intact during quantization/sparsification." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper presents a new approach for understand VLM predictions, especially relating to their robustness to perturbation. Their metric (FI) essentially estimates the change in the model output with respect to input (or parameter) perturbations. The authors test their metric by identifying for a range of images the pixels that affect model predictions the most and altering them. Furthermore, the authors test their approach with respect to input parameters by identifying crucial parameters that should be left intact during quantization, and validating that performance deteriorates less when they're not changed." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The paper does not compare against existing baselines. \n- On the sensitivity to input pixels, how does this approach compare to Grad-CAM [1] and subsequent work? It is important to see a quantitative analysis.\n- On the sensitivity to model parameters, it would be nice to see a comparison with existing approaches, e.g., [2], ...\n- I feel there is too much going on in the paper: merging sensitivity to input images + parameters at the same time seems too much for a single project. I would suggest focusing on one and studying it in detail.\n- Sensitivity of VLMs under different prompts is interesting but requires further analysis especially as to which changes in the prompts affect the influences, the semantic closeness of images and prompts, etc. \n\n\n[1] Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization, Selvaraju et al., 2016.\n[2] Decomposing and Editing Predictions by Modeling Model Computation, Shah et al., 2024." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024crack,\ntitle={Crack in the Armor: Universal Stability Measurement for Large Language Models},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1CRu6bGx25},\nnote={under review}\n}" }, "abstract": { "value": "Large Language Models (LLMs) and Vision Language Models (VLMs) have become essential to general artificial intelligence, demonstrating impressive capabilities in task understanding and problem-solving. The real-world functionality of these large models critically depends on their stability. However, there is still a lack of rigorous studies examining the stability of LLMs when subjected to various perturbations. \nIn this paper, we aim to address this gap by proposing a novel influence measure for LLMs. This measure is inspired by statistical methods grounded in information geometry, offering desirable invariance properties. Using this framework, we analyze the sensitivity of LLMs in response to parameter or input perturbations. \nTo evaluate the effectiveness of our approach, we conduct extensive experiments on models of varying sizes, from 1.5B to 13B parameters. The results clearly demonstrate the efficacy of our measure in identifying salient parameters and pinpointing vulnerable areas of input images that dominate model outcomes. Our research not only enhances the understanding of LLM sensitivity but also highlights the broad potential of our influence measure in optimizing models for tasks such as model quantization and model merging." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Large Language Models", "sensitivity analysis", "local influence measure" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/fc26c5826bd765c826a0c9ba426c7c5f46dcc26f.pdf" }, "presentation": null, "primary_area": { "value": "foundation or frontier models, including LLMs" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Crack in the Armor: Universal Stability Measurement for Large Language Models" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "$I(C_{i,q,m})$ in Eqn.(1) is not explained." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1) The paper is well-structured and easy to follow.\n\n2) The paper thoughtfully considers building strong domain-specific VLM benchmarks while sparing human annotation costs. I agree that picking the right tasks is challenging.\n\n3) Building benchmarks on existing ones with re-annotations is a smart and efficient way to control data quality and diversity. The data curation pipeline may be helpful to the community.\n\n4) Extensive evaluation results are provided. Some observations are somewhat interesting." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes a new paradigm for Vision-Language Models (VLMs) by creating multiple tasks from a single existing task, called task augmentation. This is achieved by re-annotating an existing benchmark with various tools for diverse purposes. The new paradigm is validated on the COCO and KITTI datasets. Extensive experiments on the created benchmarks are conducted, giving several interesting observations." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1) Although the idea is smart, the applicability of the data re-annotation pipeline is unknown. Currently, it is demonstrated on COCO and KITTI where instance-level annotations are provided. It would be good to elaborate more about how to generalize the data generation pipeline.\n\n2) I do not make it clear how the proposed approach can address the challenges listed in Sec.1: domain-specific validation, picking the right tasks, balancing quantity and quality.\n\n3) The notes drawn from the evaluation results seem not new for authors. Similar conclusions can be seen in various VLM evaluation papers. \n\n4) I do not see a reason why the proposed approach can be more useful than existing evaluation benchmarks. A detailed comparison with existing ones should be presented.\n\n5) The paper lacks an analysis of the evaluation results or evaluation approach." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. Could you offer a detailed demonstration of the human effort required in each dataset creation stage? This would help in understanding the resource-efficiency and automation of the \"Automatic task augmentation\" technique.\n2. How does this benchmark compare to existing VLM benchmarks in terms of task quantity, question diversity, and problem difficulty? A thorough comparison would highlight the benefits of the proposed task augmentation method.\n3. Can you clarify the task generation method using metadata? Is this done through pre-set question templates, generated by LLMs, or manual writing? A clear description of this would be valuable for reproduction. \n4. Could you include the statistical data about the 25 tasks, such as the number of questions in each task?\n\n[1] Luo Z, Xu C, Zhao P, et al. Wizardcoder: Empowering code large language models with evol-instruct[J]. arXiv preprint arXiv:2306.08568, 2023.\n[2] Muennighoff N, Liu Q, Zebaze A, et al. Octopack: Instruction tuning code large language models[J]. arXiv preprint arXiv:2308.07124, 2023.\n[3] Shypula A, Madaan A, Zeng Y, et al. Learning performance-improving code edits[J]. arXiv preprint arXiv:2302.07867, 2023." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "The paper presents a new benchmark for evaluating VLMs, which contributes for the development of this field." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents a domain-specific benchmark for evaluating Vision-Language Models (VLMs), utilizing a task augmentation technique. The benchmark provides interesting conclusions, such as considerable model performance variations across related domains. However, the primary contribution—the automatic and efficient task augmentation technique—warrants further examination. And some important details concerning the benchmark lack clarity. In summary, I think this work makes a valuable contribution but requires further revisions for publication." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The core contribution, \"Automatic task augmentation\", as claimed in line 98, appears not to be \"automatic\" nor generally available. The dataset creation still involves considerable human efforts, including metadata annotation, rule-writing, task template design, and multi-round refinement of prompts (lines 308-309).\n2. The concept of \"Task Augmentation\", although presented as new, has been thoroughly studied in previous works [1,2,3]. These works have explored methods of generating additional tasks using metadata or simple tasks for either model evaluation or instruction tuning." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "see above" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "* The paper addresses a critical issue: developing evaluation datasets for domain-specific benchmarking of VLMs\n\n* It includes an extensive evaluation using a diverse set of VLMs across various model sizes, enhancing the robustness of the findings.\n\n* The method demonstrates effectiveness, as even powerful models struggle with some tasks, demonstrating that the generated benchmark is challenging. \n\n* Human validation is incorporated to ensure clarity of image-question pairs and reduce ambiguity." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents a method for repurposing existing vision datasets to new visual tasks by leveraging the same imagery and obtaining additional metadata through a combination of human input, simple heuristic rules, and pre-trained models (e.g., segmentation and depth models). The generated data is then used to evaluate a comprehensive set of existing VLMs." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "* While the authors formalized a pipeline for “task augmentation,” the concept of repurposing existing imagery from available datasets and leveraging metadata (using off-the-shelf models or human input) to evaluate different tasks or augment training sets is well-explored in prior work. For instance, see [1],[2],[3],[4] among many others. In a way or another those benchmark repurpose existing vision datasets and use either humans or off-the-shelf models to generate additional metadata or VQA type questions. \n\n* The paper initially frames itself as a method for generating validation data for domain-specific foundation models with predefined, specific purposes. However, most models evaluated are “generalist” VLMs rather than “specialist” models. This is fine but the motivation and message should be adjusted accordingly. Additionally, while the motivation includes applications in fields like pathology and autonomous driving, no data or model relevant to these high-stakes areas is evaluated. Thus, the suitability of the pipeline for evaluating such specialized tasks remains uncertain.\n\n* The writing could be further refined, as some sections take longer to convey main points. Streamlining sections such as the introduction, Section 2.2, and Section 3.3 could improve clarity and flow.\n\n* While the proposed metric evaluation may be intuitive to the authors, incorporating more widely recognized metrics alongside individual scoring for each task could improve the benchmarks' accessibility and broader adoption.\n\n* Some important figures, like Figure 4, are difficult to interpret due to crowding. Grouping models by parameter count or model family could help clarify these visuals. Models differing in parameter count by more than 10x may not need to be displayed together unless a significant point is being illustrated.\n\n* In addition to releasing the code, sharing the final generated dataset could enhance its utility for the community, potentially offering greater practical value than the code alone.\n\nOverall, I recommend that the authors improve the writing and presentation, with an emphasis on the benchmark and findings as the main focus rather than the data generation pipeline.\n\n[1] Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs\n[2] SpatialRGPT: Grounded Spatial Reasoning in Vision Language Models\n[3] Reasoning Paths with Reference Objects Elicit Quantitative Spatial Reasoning in Large Vision-Language Models\n[4] Omni3D: A Large Benchmark and Model for 3D Object Detection in the Wild" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We introduce a task augmentation framework using metadata to create resource-efficient, domain-specific benchmarks for vision-language models, revealing that model performance varies significantly across domains, even on the same tasks." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024domainspecific,\ntitle={Domain-specific Benchmarking of Vision-Language Models: A Task Augmentation Framework Using Metadata},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1CeIRl147S},\nnote={under review}\n}" }, "abstract": { "value": "The reliable and objective evaluation of AI models is essential for measuring scientific progress and translating methods into practice. However, in the nascent field of multimodal foundation models, validation has proven to be even more complex and error-prone compared to the field of narrow, task-specific AI. One open question that has not received much attention is how to set up strong vision language model (VLM) benchmarks while sparing human annotation costs. This holds specifically for domain-specific foundation models designed to serve a predefined specific purpose (e.g. pathology, autonomous driving) for which performance on test data should translate into real-life success. Given this gap in the literature, our contribution is three-fold: (1) In analogy to the concept of data augmentation in traditional ML, we propose the concept of task augmentation - a resource-efficient method for creating multiple tasks from a single existing task using metadata annotations. To this end, we use three sources to enhance existing datasets with relevant metadata: human annotators (e.g. for annotating truncation), predefined rules (e.g. for converting instance segmentations to the number of objects), and existing models (e.g. depth models to compute which object is closer to the camera). (2) We apply our task augmentation concept to several domains represented by the well-known data sets COCO (e.g. kitchen, wildlife domain) and KITTI (autonomous driving domain) datasets to generate domain-specific VLM benchmarks with highly reliable reference data. As a unique feature compared to existing benchmarks, we quantify the ambiguity of the human answer for each task for each image by acquiring human answers from a total of six raters, contributing a total of 162,946 human baseline answers to the 37,171 tasks generated on 1,704 images. (3) Finally, we use our framework to benchmark a total of 21 open and frontier closed models. Our large-scale analysis suggests that (I) model performance varies across domains, (II) open models have narrowed the gap to closed models significantly, (III) the recently released Qwen2 72B is the strongest open model, (IV) human raters outperform all VLMs by a large margin, and (V) many open models (56\\%) perform worse than the random baseline. By analyzing performance variability and relations across domains and tasks, we further show that task augmentation is a viable strategy for transforming single tasks into many and could serve as a blueprint for addressing dataset sparsity in various domains." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "VLM", "Benchmark", "Annotation", "Ambiguity" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/a4eb7a860608608f42edb1f867e2e28e251a875a.pdf" }, "presentation": null, "primary_area": { "value": "datasets and benchmarks" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": { "value": "/attachment/4c1e34711d39d97836414add3f633b29a35adb27.pdf" }, "title": { "value": "Domain-specific Benchmarking of Vision-Language Models: A Task Augmentation Framework Using Metadata" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Please check the weaknesses and respond to the comments. Here is a summary:\n\n(1). Address the unsupported claims in the paper.\n\n(2). Include more experimental results for ablation studies, more neural architectures, and more tasks." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "This paper proposes a new activation function for deep neural networks. This is an important topic, considering the significant impact of activation function choice on deep neural network performance. The paper is clear and easy to follow." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This work presents the new logarithmic linear unit (LogLU) activation function for deep neural networks.\nThe LogLU activation solves the problem of vanishing gradient.\nThis paper shows that LogLU outperformed the other activation functions considered." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "This paper does not support some of its claims with enough evidence. For example: Under the abstract, we have: \"Its capability to solve fundamental yet complex non-linear tasks, such as the XOR problem, with fewer neurons demonstrates its efficiency in capturing non-linear patterns\". There is no evidence to support the claim that LogLU uses fewer neurons. You can strengthen this claim by providing the evidence to support this.\n\nUnder the conclusion, we have: \"The empirical results show that LogLU consistently outperforms traditional activation functions in terms of convergence speed, stability, accuracy, and loss reduction.\". The measure of stability is not clear in this paper. You can strengthen this by explaining how you observed the stability of the networks.\n\nThe experiments are limited and insufficient to conclude that LogLU is better than the other activation functions for deep neural networks. This paper did not address possible interaction with other components of a neural network (For example: dropout, learning rate, batch normalization, and so on). Please consider an ablation study that examines LogLU's interaction with other neural network components like dropout, batch normalization, etc.\n This work only considered some image classification tasks. This is not representative enough to generalize over all deep neural networks. For example, consider other cases such as simple generative models, language-based tasks, and so on." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1.\tSome inconsistent/undefined concepts? The loss function used in Section 3.2 seems to be binary cross entropy loss. While this might be obvious to some, the loss function was not defined prior to Section 3.2, which make the further discussion confusing. In Section 5, the authors talk about achieving greater “stability” with LogLU. Stability in what? This term in not (well-)defined in the paper. \n\n2.\tLack of error analysis/multiseed runs: The work lacks any error analysis (no error bars in plots or tables) whatsoever. Moreover, all the loss/accuracy curves were evaluated for a single seed. Showing the robustness of LogLU in a multiseed setting will enhance the efficacy of the proposed approach. \n\n3.\tExtend empirical comparison scope: Include additional activation functions, particularly the newer ones like Parametric RSigELU, ErfReLU, etc. to establish a more comprehensive benchmarking framework. Further, investigate LogLU’s performance on diverse and prominent architectures like DenseNet, ResNet, VGG, etc. to reinforce its general applicability.\n\n4.\tDetailed computational complexity analysis: A more granular breakdown of the time complexity will enhance the results of the paper. It might be worth performing time-complexity analysis for images instead of multiple realizations of a large vector of fixed size. Test and report the computation time of LogLU within different network architectures (e.g., shallow networks, ResNet, VGG) and layer types (e.g., dense layers vs. convolutional layers). This analysis can reveal how the activation function’s computational demands vary with the network’s depth, type, and layer configuration, especially for architectures optimized for speed.\n\n5.\tComparison to other methods for mitigating vanishing/exploding gradients issue: There are other successful and competitive methods for mitigating vanishing/exploding gradient problems at the architectural level such as the ResNet architecture. These tackle the gradient issue via architectural design using skip-connections and identity mapping to reformulate the CNN layers for learning residual functions, while specially engineered activation functions address it via their mathematical properties like non-saturating properties (LeakyReLU), gradient preservation (Swish, GELU) for negative inputs, incorporating learnable parameters (Parametric ReLU or PReLU) etc. While exploring architecture vs. activation function for solving gradient issue is out of the scope of this work (which focuses solely on activation functions), a detailed discussion highlighting other non-activation function based techniques for overcoming vanishing/exploding gradient problem will help with the completeness of the paper. \n\n6.\tExamine gradient flow in various conditions: Explore gradient dynamics with respect to learning rate schedules and optimizers to provide insight into how LogLU performs under different training regimes. Additionally, ablation studies on placement within specific layers could clarify LogLU’s most impactful applications.\n\n7.\tTheoretical insights on regularization effect: Since the logarithmic component potentially regularizes activations for negative inputs, discussing theoretical implications related to regularization could open new perspectives on the theoretical advantages of LogLU in avoiding overfitting." }, "rating": { "value": 1 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. Innovation in activation functions: The proposal of LogLU as a hybrid activation function is novel and provides an interesting alternative to traditional activation functions. The logarithmic component for negative inputs introduces a unique way to handle the dead neuron problem while also limiting gradient vanishing, especially compared to ReLU and Leaky ReLU.\n\n2. Experiments performed: The authors performed evaluations on classification benchmark datasets (Caltech 101 and Imagenette) and used InceptionV3 architecture for the classification task. The consistent improvements in Val accuracy (on Caltech 101) and convergence speed were presented in Tables 3 and 4 and Figures 4 and 5, which suggest that LogLU might be a competitive alternative to existing activation functions.\n\n3.\tPerformance on classification task: By demonstrating that LogLU can solve the XOR problem with a simplified architecture, the authors underscore LogLU’s efficiency in capturing non-linear relationships with fewer neurons, an advantage for both resource efficiency and model scalability.\n\n4.\tAddressing gradient problems: The paper discusses how LogLU mitigates the vanishing and exploding gradient problems, which are common in deeper networks due to the use of traditional activation functions. LogLU’s bounded gradient across all input values is well-explained and experimentally supported, potentially making it an optimal choice for complex neural architectures.\n\n5.\tEfficient computation: The paper also presents an analysis of computation times, demonstrating that LogLU is computationally efficient (Figure 2). LogLU achieves an average computation time significantly lower than other activation functions (except ReLu and Leaky ReLU), with performance that consistently outpaces more complex alternatives like Mish and Swish." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper introduces a new activation function, Logarithmic Linear Unit (LogLU), aimed at addressing issues inherent in widely used activation functions like ReLU, Leaky ReLU, ELU, etc. LogLU uses a logarithmic function for negative inputs, allowing it to maintain active gradients even with negative inputs, potentially reducing issues like dead neurons and vanishing gradients. Experiments are conducted comparing LogLU with other established activation functions across datasets like Caltech 101 and Imagenette using the InceptionV3 architecture. The authors highlight benefits in convergence speed and accuracy, proposing LogLU as a robust alternative for deep learning models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.\tLack of a rigorous study/analyses: Although the paper tries to solve an important problem in deep learning based training of CNNs in the presence of vanishing/exploding gradient problem, the work done in the current version of the paper appears to be very preliminary in nature and there is a huge scope for improvement. \n\n2.\tComparison with more recent activation functions: While the paper covers popular functions like ReLU, ELU, and Swish, it could benefit from comparisons with other activation functions such as SiLU, GELU, Softplus or more recent alternatives like Parametric RSigELU (Kiliçarslan, et al, Feb 2024) and ErfReLU (Rajanand, et al, May 2024). Including such comparisons would provide a broader perspective on LogLU’s competitive positioning.\n\n3.\tAccuracy on Imagenette dataset: There does not seem to be any significant gain in performance on the Imagenette dataset, where activations such as Swish and Mish marginally beat the proposed activation function. Therefore, the claims of better performance is not applicable on this dataset. \n\n4.\tComputational Complexity Analysis: Although the authors claim computational efficiency, the complexity analysis could be strengthened. The time complexity is presented in aggregate form (average time over multiple runs), but there is limited discussion on LogLU's computational demands relative to exponential or polynomial components in activation functions like ELU or Mish, which could help enhance the claims of efficiency.\n\n5.\tScalability to other deep CNNs and datasets: While the experiments are valuable, they focus primarily on moderately sized datasets for only image classification tasks. Testing LogLU on larger datasets, such as the MNIST, CIFAR10, COCO, CelebA, Pascal VOC, SVHN, etc., and using architectures beyond InceptionV3 (e.g., ResNet or transformer-based models) could provide deeper insights into LogLU’s applicability in large-scale settings.\n\n6.\tScalability to loss functions beyond cross-entropy: Since the gradient computation depends on loss function, it would be highly valuable to assess the effectiveness of LogLU for different loss functions for the classification task. These directions were not explored in the current version of the work. \n\n7.\tScalability to tasks beyond classification: The effectiveness of LogLU on other tasks such as image segmentation, object detection or image generation, etc. remains unexplored. The work could potentially benefit from showing superior performance/computational efficiency over other activation functions in a variety of other prominent computer vision tasks.\n\n8.\tAblation Studies: The effectiveness of LogLU in specific neural network layers (e.g., convolutional layers vs. dense layers) or different learning rates and optimizers remains unexplored. Adding ablation studies could help isolate the benefits of LogLU more distinctly across various configurations." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 2 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "* On page 1, the manuscript states: \"Although Leaky ReLU addresses this problem by permitting small negative values, it introduces the vanishing gradient problem, limiting its effectiveness in deep networks (Maas, 2013).\" However, I believe that Leaky ReLU does not introduce the vanishing gradient problem. In fact, Leaky ReLU was proposed to mitigate issues like the dying ReLU problem by allowing a small, non-zero gradient for negative input values. Additionally, no such discussion regarding Leaky ReLU introducing vanishing gradients is found in Maas et al. (2013).\n\n* On page 5, the manuscript states that Table 1 shows the derivative of LogLU, but Table 1 does not include this information. Please update Table 1 to include the derivative of LogLU or revise the manuscript to accurately reflect the contents of Table 1.\n\n* On page 6, the term \"more controlled activations\" is ambiguous and requires clarification. The authors should provide a clear definition or explanation of what is meant by \"more controlled activations\" to enhance the reader's understanding.\n\n* The lines in the figures are difficult to distinguish. Please use more distinct colors or linestyles to enhance clarity. \n\n* On page 7, why are the model sizes different across datasets, even though Inception-V3 is used for both?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "* The proposed LogLU is very simple while being both continuous and differentiable. It requires less computation than modern activation functions such as Swish because it does not involve exponential computations in either the forward or backward pass, although it only requires logarithmic computation in the forward pass." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes LogLU as a new activation function, which is both continuous and differentiable.\nLogLU is empirically shown to be computationally more efficient compared to modern activation functions such as Swish or Mish, but requires slightly more computation than ReLU or Leaky ReLU.\nThe authors claim that a simple one-hidden-layer MLP with LogLU activation can learn the XOR function.\nLogLU is compared to other activation functions using the Caltech-101 and Imagenette (a simplified variant of ImageNet) datasets with the Inception-V3 architecture, demonstrating faster convergence of models with LogLU activation." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "* The authors claim that a simple MLP with LogLU activation can learn the XOR function, highlighting this as an advantage of using LogLU. However, MLPs with other activation functions are also capable of learning the XOR function. The authors should discuss why using LogLU is more advantageous than other activation functions in the context of the XOR example.\n\n* The experimental evaluations are insufficient. At a minimum, it is necessary to compare the proposed activation function with other methods using network architectures beyond Inception-V3. Additionally, each experiment should be conducted with various random seeds to assess the variability of the outputs (loss or accuracy)." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. How to justify the theoretical reason for using the log function, could you give any intuition?\n2. Here are some thoughts to justify LogLU and address the theoretical side. $f=-\\log(-x+1)$ solves an instance of Monge-Ampère equation $$\\log\\det f''=2f$$ \nwhere $\\det$ is the analog in the high-dimensional case, associated with Dirichlet boundary condition $\\lim_{x\\to\\partial \\Omega} f=\\infty$ on the domain $\\Omega=(-\\infty,1)$. We can alternatively set a Neumann boundary condition $f'(0)=1$ on $\\Omega=(-\\infty,0)$ to guarantee the $C^1$ continuity. The intuition is that the logarithmic curvature is proportional to the value. The property includes self-concordance and logarithmic homogeneity.\nSee [1] in Chapter 2.3.3: properties; Chapter 2.5: universality---the log function as a canonical construction.\nSee [2] in Proposition 1.4.3: a connection with the Calabi theorem.\n\n[1] Interior point polynomial time methods in convex programming. A. Nemirovski 2004.\n\n[2] Conic optimization: affine geometry of self-concordant barriers and copositive cones. R. Hildebrand 2017." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. I believe this direction of activation search is fundamentally impactful in deep learning, because it changes the basic part of neural networks. Though the experiments are very limited, it is already a good sign that this important part works better.\n2. The paper's message is minimal, direct, and clear." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes to use $-\\log(-x+1)$ in ReLU instead of $0$ when the input is negative. The performance on fine-tuning InceptionV3 on Caltech101 and Imagenette is improved over ReLU, ELU, Leaky ReLU, Swish (SiLU) and Mish." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. My major concern is that the experiments are restricted to very limited data and models, so LogLU's validity is still questionable on other models and tasks.\n2. More specifically, the results would be convincing if the author could add experiments on common models, such as , ResNet, UNet, and Transformers. If LogLU works on more models I believe it will improve the paper. \n3. Another solution that could help is to ask if it is possible to find a dataset or a toy model where LogLU significantly outperforms other activations.\n3. The model has 73M parameters for Caltech 101 and 37M for Imagenette, both pre-trained on the Imagenet dataset. I don't understand why the models are both InceptionV3 but are different in size.\n4. I don't understand why the experiments only include fine-tuning, but not training from scratch." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "The Logarithmic Linear Unit (LogLU) is a novel activation function designed for deep neural networks, improving convergence speed, stability, and overall model performance" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024logarithmic,\ntitle={Logarithmic Linear Units (Log{LU}s): A Novel Activation Function for Improved Convergence in Deep Neural Networks},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1D3TjFidCS},\nnote={under review}\n}" }, "abstract": { "value": "The Logarithmic Linear Unit (LogLU) presents a novel activation function for deep neural networks by incorporating logarithmic elements into its design, introducing non-linearity that significantly enhances both training efficiency and accuracy. LogLU effectively addresses common limitations associated with widely used activation functions include ReLU, Leaky ReLU, and ELU, which suffer from issues like the dead neuron problem and vanishing gradients. By enabling neurons to remain active with negative inputs and ensuring effective gradient flow during backpropagation, LogLU promotes more efficient convergence in gradient descent. Its capability to solve fundamental yet complex non-linear tasks, such as the XOR problem, with fewer neurons demonstrates its efficiency in capturing non-linear patterns. Extensive evaluations on benchmark datasets like Caltech 101 and Imagenette, using the InceptionV3 architecture, reveal that LogLU not only accelerates convergence but also enhances model performance compared to existing activation functions. These findings underscore LogLU's potential as an effective activation function that improves both model performance and faster convergence." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Activation Function", "Deep Neural Networks", "Optimisation" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/daa47b24c9ed168fdcd48a71eb55f57188cf2b30.pdf" }, "presentation": null, "primary_area": { "value": "learning theory" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Logarithmic Linear Units (LogLUs): A Novel Activation Function for Improved Convergence in Deep Neural Networks" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": { "value": "Thank you for the review. We address your comments below.\n\nWeaknesses:\n- \"The lunch menu optimization example, while a clear illustration, does not really motivate the problem from a practioner's perspective\"\n\nAgreed, we chose this for simplicity. However, utility maximization is an extremely natural problem. Other examples for utility maximization include: adding medical services to a healthcare package as to maximize the welfare of all patients, adding features to a website to maximize user engagement, etc...\n\n- \"There are no p-system experiments\"\n\nIndeed, we couldn't find a real-world dataset for this problem. Previous papers seems to either be completely theoretical (no experiments), or run experiments just under a cardinality constraint.\n\nQuestions:\n\nQ1) It is quite natural in welfare maximizations (e.g., many people with different preferences). Another example is finding a representative set of images (e.g., thumbnails for a video). Here N can be very large (the number of frames in the video), clearly there is plenty of redundancy, so our approach is very natural here.\n\nQ2) It is defined just above Model 1. It is the set $\\{f^i(e)\\}_{i\\in [N]}$ and in our models we assume that every $f^i(e)$ (the value of the i-th func on e) is a random variable, not the set $A_e$.\n\nQ3) We roughly followed the paper of Rafiey and Yoshida which introduced this set. They simply say that they select a set of \"popular pickup locations in the dataset\". We used k-means to pick \"popular locations\"." }, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": { "value": "Thank you for your review. \n\nAbout the figure, basically all algorithms achieve almost the same quality when we sample enough elements.\nWe will move Section 4 before Section 3 if accepted." }, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": null, "comment": { "value": "Thank you for your review. \n\nWe would like to emphasize that our main contribution is the introduction of the *uniform* sampling algorithm, observing that it outperforms other approaches empirically, and using smoothed analysis to bridge the gap between theory (no worst case analysis possible) and practice. We believe that this algorithm can be used as the first line of attack for many real-world massive datasets. The improved weighted sampling is \"nice to have\" and lays the groundwork for the smoothed analysis of the uniform sampling algorithm, but this is not our main contribution.\n\nWe address your questions below. \n\nQ1) An upper bound is not sufficient. This is because our proofs require a *multiplicative* error bound for the minibatch approach to work. Consider the following example: all functions except one are always zero ($f^i \\equiv 0, i\\neq j$), and one function, $f^j$, is upper bounded by 1. Clearly both the minibatch and the sparsifier algorithms can't optimize the sum as they will keep sampling functions that are always 0. The above example is unlikely to appear in real world applications, but it illustrates that worst-case analysis is simply not the right tool here. This is why we use smoothed analysis to explain the superior performance of uniform sampling in practice.\n\nQ2) Yes, specifically Model 2. The assumptions of Model 2 are *extremely* mild and we verify empirically that they hold for *all* of our datasets. We would like to emphasize that we only introduced smoothed analysis in this revision of the paper, while we used the same datasets in previous revisions. That is, we did not simply pick datasets where our models apply (and indeed Model 1 does not apply to all datasets)." }, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": null, "primary_area": null, "questions": null, "rating": null, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": null, "summary": null, "supplementary_material": null, "title": { "value": "Our main contribution is the uniform alg + smoothed analysis." }, "venue": null, "venueid": null, "weaknesses": null, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "* If the $f_i$ are all bounded by a value $R$, could theoretical guarantees be gotten for uniform sampling?\n* Do you expect Models 1 and 2 would hold widely in applications of decomposable functions?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Exploring submodular optimization algorithms that do not view the function $f$ as simply a black box is an interesting research direction that I think deserves attention.\n- They explained their results clearly and the paper was easy to understand." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper considers maximization of decomposable monotone submodular functions over a ground set of size $n$, meaning that the objective function $f$ is a sum of $N$ monotone submodular functions $f_1,...,f_N$. If $N$ is large, then evaluations of $f$ may be computationally demanding. Previous work on the topic (Rafiey & Yoshida, 2022; Kenneth & Krauthgamer 2023) proposes constructing a random sparsified version of $f$ that is a weighted sum of some subset of the functions, and is within a multiplicative $\\epsilon$ factor approximation on all sets. A sparsifier such as those mentioned could be constructed as a preprocessing step for an algorithm, and then the algorithm would be run using the sparsifier in place of the original function. The state of the art is that of Kenneth & Krauthgamer, where a sparsifier of $O(k^2n\\epsilon^{-2})$ functions is constructed using $O(Nn)$ oracle calls. The sparsifier is constructed by iterating over the functions, computing a probability $p_i$ for each function $f_i$ to be included, and then sampling that function with probability $p_i$ (which takes a total of $O(Nn)$ queries). Then querying the sparsifier takes $O(k^2n\\epsilon^{-2})$ function evaluations, compared to $O(N)$ function evaluations to query the original $f$. If $N$ is relatively large, the sparsifier is more efficient.\n\nInstead of computing a sparsifier as a preprocessing step for an algorithm, this paper proposes a \"mini-batch method\" (which have been used in other areas of ML) for this problem (Algorithm 3). That is, a new sparsifier is sampled every iteration of the greedy algorithm. The approach in this paper uses the same sampling probabilities $p_i$ as Kenneth & Krauthgamer, and therefore still needs the $O(Nn)$ queries as a preprocessing step to compute the $p_i$. In order to prove some of the results in their paper, they make additional assumptions on the problem setting (Models 1 and 2). Several analyses are done on the number of function queries needed for their algorithm. Finally, they include an experimental comparison of their algorithm and related works." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- It seems a lot of the difficulty of these sparsification approaches is because the sampling of the $f_i$ is non-uniform, but it is still unclear to me that this is so much better than uniform sampling. According to this paper, uniform sampling does better in practice, and requires no preprocessing to compute the $p_i$ since they would be uniform. It is also stated that no theoretical bound can be gotten for uniform sampling. But if we assume that all the $f_i$ are bounded by some value $R$, why can't concentration inequalities be used to get a theoretical guarantee for the uniform approach?\n- Some of the results are dependent on assuming Models 1 or 2 (see Table 1), but it isn't clear to me that these models are realistic for applications of the problem.\n- Improvements over Kenneth and Krauthgamer mainly include the curvature of the function in the bound on the number of function queries, so the bounds are instance dependent.\n- The bounded curvature results (which don't depend on Models 1 and 2) don't use ideas that are that novel compared to related work. It seems the biggest difference from Kenneth and Krauthgamer is computing the sparsifier at each round of the greedy algorithm, and only relatively minor changes are needed to the argument of Kenneth and Krauthgamer." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "It might be better to put Section 4 before Section 3 to ensure the continuity of the analysis." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "Overall, the paper is well-structured and easy to understand. The definitions and explanations are clear, and related work is discussed in sufficient detail. \n\nThe discussion on uniform and weighted sampling, along with the smoothing model, helps bridge the gap between theoretical results and the empirical performance of the algorithms. It provides insights into why an algorithm without a worst-case guarantee can still perform well in experiments." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper addresses the problem of maximizing a non-negative, monotone, decomposable submodular function under the cardinality constraint and $p$-system constraint. It introduces the first mini-batch algorithm with weighted sampling for this problem, demonstrating that it outperforms the sparsifier-based approach both theoretically and empirically. Additionally, the authors observe that, in experiments, uniform sampling outperforms weighted sampling. To explain this outcome, they define two smoothing models. The first model provides theoretical guarantees for both the mini-batch and sparsifier algorithms on some datasets, while the second model applies only to the mini-batch algorithm but is effective across all datasets tested." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "The algorithm is simple, and the analysis is quite straightforward. The technical contribution is limited. \n\nWith 12 indistinguishable lines in Figure 1, it is hard to see which algorithm with $\\beta=10^{-2}$ achieves the best performance." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- In the introduction, you claim that \"in many of the above applications, $N$\n (the number of underlying submodular functions) is extremely large, making the\n evaluation of $F$ prohibitively slow.\" Are there realistic examples where $N\n \\gg 1000$? It's not clear to me how often we really encounter $N$ *distinct*\n personalized submodular functions.\n- What exactly is the quantity $A_e$ when you first introduce it on page 3?\n This should be made more clear. Initially, I thought it was a vector of all\n marginal values, but then in model 1 you say it's a random variable.\n- For the Uber pickups experiment, why do you use Llyod's algorithm to find\n centers instead of a data-indepedndent grid?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Uses smoothed analysis to more accurately study realistic inputs\n- Table 1 cleanyl describes the results, including a comparison with [Kenneth-Krauthgamer, ICALP 2024]\n- Draws connections to the lazier-than-lazy greedy algorithm of [Mirzasoleiman et al., AAAI 2015]\n and explains how the two ideas can be combined to reduce query complexity by a factor of $\\Theta(k)$\n- Good comprehensive set of experiments for cardinality constraints, though the\n values of $k \\le 20$ are quite small. It would be nicer to increase $k$ to see\n how fast the different algorithms converge (relatively) to lazy greedy" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This work studies a sampling-based algorithm for faster non-negative monotone *decomposable* submodular maximization subject to\ncardinality or $p$-system constraints. In particular, it builds on work of\n[Kenneth-Krauthgamer, ICALP 2024] (please update reference in paper), which sparsifies\nand reweights the set of functions $f^{(i)}(S)$ for the input function $F(S) = \\sum_{i=1}^N f^{(i)}(S)$.\nThe goal of this paper is to eliminate the dependence on $N$, which the authors do under mild assumptions\nvia *smoothed analysis*. They also show that this is not possible in the general case with a simple pathological example.\nIn short, the main idea is to sample a subset of $f^{(i)}(S)$ functions at each step to form a\n\"mini-batch\" for approximating the full $F(S)$. The algorithm then greedily\nselect the next element based on the sampled funciton (which changes in each iteration), not $F$ itself.\n\nFurther, under the mild realistic assumptions, they prove why uniform sampling is a competitive approach,\nwhich helps explain initially surprising experimental observations.\nLastly, this work provides a clean set of experiments comparing their mini-batch sampling-based methods to\na full lazy greedy algorithm and the sparsification idea in [Kenneth-Krauthgamer, ICALP 2024]." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- The lunch menu optimization example, while a clear illustration, does not\n really motivate the problem from a practioner's perspective\n- There are no $p$-system experiments\n- It is unclear if ICLR is an appropriate venue for this work. The\n non-exhaustive list of topics in the Call for Papers includes \"optimization\",\n but submodular maximization in its raw form seems one hop away from the\n target areas of ICLR (deep learning)" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "Present the first mini-batch algorithm for submodular maximization; use smoothed analysis to justify performance" }, "_bibtex": { "value": "@inproceedings{\nanonymous2024minibatch,\ntitle={Mini-batch Submodular Maximization},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1DEEVAl5QX},\nnote={under review}\n}" }, "abstract": { "value": "We present the first *mini-batch* algorithm for maximizing a non-negative monotone *decomposable* submodular function, $F=\\sum_{i=1}^N f^i$, under a set of constraints. \nWe consider two sampling approaches: uniform and weighted. We show that mini-batch with weighted sampling improves over the state of the art sparsifier based approach both in theory and in practice. Surprisingly, we experimentally observe that uniform sampling achieves superior results to weighted sampling. However, it is *impossible* to explain this using worst-case analysis. Our main contribution is using *smoothed analysis* to provide a theoretical foundation for our experimental results. We show that, under *very mild* assumptions, uniform sampling is superior for both the mini-batch and the sparsifier approaches. We empirically verify that these assumptions hold for our datasets. Uniform sampling is simple to implement and has complexity independent of $N$, making it the perfect candidate to tackle massive real-world datasets." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "smoothed analysis", "submodular maximization" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/644548452ada9eb3f77b2c0cf6d9d7a2258281e0.pdf" }, "presentation": null, "primary_area": { "value": "optimization" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Mini-batch Submodular Maximization" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Will there be a significant performance difference if the model is not pre-trained with UMT?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The method have better performance than baselines without additional cost." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper addresses a solution for better long-form video understanding using a method named Adaptive Memory Mechanism (AMM). This method enables the Vision Transformer (ViT) to adjust its temporal receptive field dynamically depending on the input video. A memory bank is utilized to save the most important Key-Value when temporally processing the videos. The proposed method is tested on AVA and Epic-Kitchens datasets for action detection, recognition, and anticipation tasks. Experiment results show performance improvement to the ViT baselines without additional cost." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The paper lacks SoTA comparisons. Is the task different from common action recognition and action detection? Multiple methods such as VideoMAE, Omnivore, or MMT have been tested on these datasets. It would be helpful if the authors could explain the difference between previous SoTAs with the proposed method, for example in parameter count or GFLOP difference.\n2. The improvement to ViT and MeMVit baselines is marginal.\n3. There is no difference in the FLOPs and Param(M) numbers compared to the baselines. Can the authors explain further the efficiency advantage achieved by the proposed method?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Please see weaknesses" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 1 }, "strengths": { "value": "* Long-form video understanding is an important video research topic and the idea of using an adaptive memory bank sounds reasonable and promising. \n* Compared to MeMViT, the results show consistent improvements though some datasets only have marginal gain." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper presents an adaptive memory method to improve the existing memory-augmented methods for long-form video understanding. The method is based on MeMViT but makes the memory bank adaptive to support the adaptive temporal receptive field. The experiments are conducted on Ava and Epic-Kitchens dataset with the comparison with ViT and MeMViT." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "* One of the main motivations of the paper is to retain embeddings instead of discarding memory when the memory limit is reached. However, based on the experiments, it's unclear if the effective receptive field of AMViT is indeed larger than MeMViT through the proposed adaptive memory module. Are they still using the same memory bank size?\n* In the model section, the paper presents two new modules, including Input-aware selective module (ISM) and Adaptive Memory mechanism(AMM). However, there are no ablations to validate the individual effectiveness of these modules.\n* How do we select parameters for MeMViT? Some parameters for MeMViT (Table 6) are not defined, e.g, memory bank size. Is it the same as AMViT? Given the authors are reproducing MeMViT with a different backbone, how the results compare to the original paper.\n* In Table 1, it's unclear why all the three methods are having the same FLOPs and parameters given MeMviT and AMViT has additional memory bank modules. It's also better to conduct run-time comparison.\n* The experiments are also missing a system-level comparison with the current SOTA results on the benchmarks." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. Does the KV Cache in this paper retain the gradient?\n2. This paper focuses on a pure vision model with enhanced memory design. However, the ViT-only architecture is capable of a limited range of video-related tasks. Is it possible to integrate it with video-language models to achieve wider range of video tasks to exert more impact on the community?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The analysis of the limited temporal receptive field in long-term video understanding makes sense, and the motivation is clear.\n2. The method is simple and intuitive." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper aims to enhance ViT for long-term video understanding. The authors design a memory bank to store historical information and develop input-aware adaptive memory selection to retrieve the relevant information to assist long-term analysis. The experiments show that the architecture demonstrates satisfactory performance with high efficiency." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The experiments are limited. Only AVA and Epic-Kitchens are reported. Results on more video datasets are required to verify the effectiveness of the adaptive memory design. Besides, the performance improvements are marginal.\n2. The memory bank is recurrently updated by adaptive selection. Is it possible that in a long video, the content in the middle of the video is not closely related to the beginning, and only relevant content appears towards the end? However, during the memory bank update process, the tokens of the earlier video content were already discarded." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "When comparing with MeMViT, your model uses the memory bank and the selected Q-V cache, while MeMViT only uses Q-V cache. Have you ensured that the number of embeddings in both model is consistent? Specifically, does the size of the memory bank plus the size of the selected Q-V cache match the size of the unselected Q-V cache?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1.Long-form video understanding is an important research task, and the author has provided a reasonable solution.\n\n2.The paper is well-written, making it easy to read." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes an Adaptive Memory Mechanism (AMM) for Vision Transformer (ViT) in long-form video understanding. It addresses the issue of selecting an optimal Temporal Receptive Field by allowing ViT to adjust TRF dynamically. Instead of directly discarding early Key-Value cache, AMM uses a Memory Bank to retain important embeddings from the Key-Value cache based on attention scores. Experiments on AVA and Epic-Kitchens show the advantages of AMM in action recognition, anticipation, and detection tasks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1.The novelty of memory bank is limited. Many studies have explored how to utilize memory to retain important historical information and how to dynamically update memory. For example, Xmem[1] prioritizes retaining the most frequently used candidates. MA-LLM[2] and MovieChat[3] merge the two most similar candidates based on similarity once the memory bank capacity is exceeded. The innovations and advantages of the memory bank proposed in this paper compared to these methods are unclear.\n\n2.The fairness of the experiment is in question. When comparing with the baseline model MeMViT, the authors replaced the backbone of MeMViT from MViT to UMT. This seems to have led to a decline in the performance of the baseline model. For example, in the EPIC-KITCHEN-100 action recognition task, the performance reported in the original paper on MeMViT was 48.4%, while the performance presented in this paper is 43.03%. The authors should maintain the same settings as MeMViT for the experiments to make the results more credible.\n\n3.The performance improvement is limited. Compared to the baseline model MeMViT, the performance improvement is less than 1% in all experiments.\n\n4.Lacks of comparison with the latest methods. This article only presents comparisons with ViT and MeMViT. Some recent methods are missing, such as MAT[4] and MC-ViT[5].\n\n5.Lacks of necessary ablation studies. (2) This paper uses an input-aware selective module to prevent redundant embeddings from being retained, and uses a memory bank to retain useful embeddings. However, there are no ablation experiments to demonstrate the effectiveness of these two components individually. (2) The lack of ablation experiments on the memory bank update method. For example, comparing the update of the memory bank using attention score of class tokens proposed in this paper with previous methods (see weakness 1) and First-In-First-Out (FIFO).\n\n[1] XMem: Long-Term Video Object Segmentation with an Atkinson-Shiffrin Memory Model, ECCV 2022\n[2] MA-LMM: Memory-Augmented Large Multimodal Model for Long-Term Video Understanding, CVPR 2024\n[3] MovieChat: From Dense Token to Sparse Memory for Long Video Understanding, CVPR2024\n[4] Memory-and-Anticipation Transformer for Online Action Understanding, ICCV 2023\n[5] Memory Consolidation Enables Long-Context Video Understanding, arxiv 2024" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "Please revise the Weaknesses section point by point. This is a paper with great potential. If the authors can provide additional responses to certain issues, discuss related work more thoroughly, and include more experiments and observations, I would be very happy to raise my score." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Long-form video understanding is an important task, and efficiency is indeed a crucial metric in this context.\n\n2. The proposed method can reduce both training and inference costs.\n\n3. Introducing a memory bank to handle long sequence inputs is intuitive and reasonable." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper introduces an Adaptive Memory Mechanism (AMM) to improve Vision Transformers (ViT) for long-form video understanding. AMM dynamically adjusts the Temporal Receptive Field (TRF) based on video content, overcoming limitations of fixed TRF approaches that either lose key information or increase computational costs. Experiments show that AMViT, integrating AMM, outperforms existing models like MeMViT in tasks such as action recognition, anticipation, and detection, while reducing computational overhead, validated on datasets like AVA and Epic-Kitchens." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. (important) The number of benchmarks (only 2) and baselines (also only 2) compared seems somewhat limited. Adding more experiments would make the paper more convincing.\n\n2. (important) Although the authors emphasize that the new architecture is designed for long-form video, this aspect is not discussed in the experimental section. Are the benchmarks presented in the paper truly for long videos, and what is the average input length? It would have been better if the authors had conducted more detailed evaluations on benchmarks like MovieChat-1K [1] or LongVideoBench [2].\n\n3. The writing and figures in the paper need improvement, especially regarding the notation for memory. There are too many subscripts and superscripts, along with the extensive use of qkv notations, which made it take me three times longer to understand the entire paper. \n\n[1] Song, Enxin, et al. \"Moviechat: From dense token to sparse memory for long video understanding.\" Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.\n\n[2] Wu, Haoning, et al. \"Longvideobench: A benchmark for long-context interleaved video-language understanding.\" arXiv preprint arXiv:2407.15754 (2024)." }, "withdrawal_confirmation": null }, { "TLDR": { "value": "We introduce the Adaptive Memory Vision Transformer (AMViT), which dynamically adjusts its Temporal Receptive Field using an Adaptive Memory Mechanism for effectiveness and efficiency improvement in long-form video understanding." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024adaptive,\ntitle={Adaptive Memory Mechanism in Vision Transformer for Long-form Video Understanding},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1DEHVMDBaO},\nnote={under review}\n}" }, "abstract": { "value": "In long-form video understanding, selecting an optimal Temporal Receptive Field (TRF) is crucial for Vision Transformer (ViT) models due to the dynamic nature of diverse video motion contents, which varies in duration and velocity. A short TRF can result in loss of critical information, while a long TRF may decrease ViT's performance and computational efficiency caused by the unrelated contents in videos and the quadratic complexity of the attention mechanism. To tackle this issue, we introduce Adaptive Memory Mechanism (AMM) that enables ViT to adjust its TRF dynamically in response to the video's dynamic contents. Instead of discarding Key-Value (KV) Cache from the earliest inference when the settings limit is reached, our approach uses a Memory Bank (MB) to retain the most important embeddings from the Key-Value Cache that would otherwise be discarded in memory-augmented methods. The selection is based on the attention score calculated between the Class Token (CLS) in current iteration and the KV Cache in previous iterations. We demonstrate that Adaptive Memory Vision Transformer (AMViT) outperforms existing methods across a diverse array of tasks (action recognition, action anticipation, and action detection)." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Key-Value Cache", "Vision Transformer", "Video Understanding" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/89a290cc5b5541967c05c287db67008301f18a46.pdf" }, "presentation": null, "primary_area": { "value": "applications to computer vision, audio, language, and other modalities" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Adaptive Memory Mechanism in Vision Transformer for Long-form Video Understanding" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "- To be clear, in these experiments, the model might not actually abstain, right? You would have to calculate these metrics and then hide the response from the user if it were deemed unacceptable, right?\n- It was hard to tell if there was a proposed training or inference method from reading the intro. It took me a while to realize that this was more of an analysis paper, showing how these metrics could be used for filtering model outputs.\n- Sec. 3 probably doesn’t need to take up as much space is it currently does (people should know what NLL is), but at the same time it could give more understanding into the metrics (computing actual entropy over samples is hard, so you compute predictive entropy).\n- L.377 typo “unanswerable vs. unanswerable”\n- L.471 “abstaining to answer”→ “abstaining from answering”" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 4 }, "strengths": { "value": "- Important: The core idea of the paper is sensible, relating uncertainty metrics to abstention in order to improve factuality and safety of model responses.\n- Important: The experiment design is sound and the chosen metrics are reasonable. The experiments include many relevant datasets for measuring knowledge, hallucination, and safety.\n- Of some importance: The paper is fairly well-written and easy to follow." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper studies how token-level and semantic uncertainty metrics on generated LLM text relate to accuracy on knowledge-intensive tasks, hallucination on unanswerable questions, and response safety on adversarial/malicious question datasets. Among the uncertainty metrics explored is one based on on counting hedge words in model responses. Experiments show that these uncertainty metrics are useful for model abstention in order to improve correctness of model generations, reduce hallucination, and increase response safety (at the cost of an increased abstention rate). Experiments are conducted across many relevant datasets using Llama 2 models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- Very important: The novelty of the work is quite limited in my view. The high level conclusion that uncertainty is useful for abstention has already been thoroughly explored. What I see as this paper’s contributions beyond this observation are: (1) measuring in-dialogue uncertainty can help with abstention, specifically for factuality/hallucination; (2) uncertainty can help with safety, as it turns out that responses on AutoDAN-like datasets are more likely to be unsafe if they are uncertain. I don’t think the paper claims much beyond this. So a further issue with the novelty here is that (1) has already been shown, more or less, in https://arxiv.org/abs/2405.21028. The (2) result is interesting but I do not think it is a large enough result for a full paper, and it is not explored in much depth beyond one paragraph in this paper.\n- Important: The measurement of in-dialogue uncertainty, even if useful, is a heuristic that does not feel particularly generalizable, especially compared to other model-based measurements of in-dialogue confidence." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 1 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- How can we practically account for all possible hedge words for every use case? Some prompts might even require responses to include hedge words; seems like a lot of finetuning and engineering effort to incorporate this uncertainty metric\n- I'm not sure I agree with hallucinations being only considered for unanswerable questions. LLMs definitely hallucinate in other situations. How extendable are these findings?\n- Statistical uncertainty metrics perform at more or less the same level. What should the reader take away from all these results?" }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Well-written paper with a clear walkthrough over the different problems and uncertainty metric considerations\n- Interesting idea of using in-dialogue uncertainty as a measure of response uncertainty\n- Clear description of experiments, metrics, and results; strong scientific method" }, "student_author": null, "submission_guidelines": null, "summary": { "value": "Paper shows that abstention based on different measures of uncertainty for different types of prompts works well. Specifically, for correctness and safety, statistical uncertainty-based abstention helps improve correctness and reduce unsafe responses. For hallucinations, abstention based on in-dialogue uncertainty (coined by authors as the inclusion of phrases such as \"I don't know\" in model responses) helps reduce hallucinations." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- It should not come as a surprise that using uncertainty metrics helps LLMs abstain when they should not engage with the prompt, as shown in Kadavath (2022) and multiple other papers cited in the related works. The core contributions of this paper can be boiled down to the introduction InDU (which also was inspired by an existing paper by Islam (2020)) and when to use each kind of uncertainty, both of which seem more fitting for, e.g., an appendix in Kadavath's paper, especially since this reads more like a survey paper of implementation details than novel ideas or concepts\n- Minor: various typos such as \"In-Dialogoe\" in Introduction, Islam et al. without year in 3.2" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 5 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "- It would be worth discussing the tradeoff between abstention and usability further. \n- In-Dialogue Uncertainty is given an acronym but the acronym isn't used. It's also misspelled on L053." }, "rating": { "value": 3 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "- **Well written and clearly organized**: the paper is easy to follow, the writing is clear, and the questions being tested are clear. \n- **In-dialogue uncertainty metric is new**: As far as I can tell, past work has not proposed counting the number of hedge words as a method of confidence estimation. \n- **Sufficient datasets examined**: The authors do a good job of testing multiple datasets to make their point." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper focuses on abstention and uncertainty in LLMs, benchmarking how useful different uncertainty estimates are across three broad tasks.\nThese tasks are correctness, unanswerable questions, and safety. \nCorrectness is evaluated against standard QA data (TriviaQA, SciQA, CoQA, StrategyQA, and GSM8K).\nUnanswerable vs answerable questions are sourced from the SelfAware dataset and SciQA. \nAdversarial examples are sourced from AttaQ and AutoDAN. \nThe authors examine negative log-likelihood, predictive entropy, semantic entropy, and In-Dialogue Uncertainty, which is the number of hedge tokens present in the output. \nAll experiments were run on Llama2.\nAcross different tasks, the authors find that different uncertainty estimates lead to better or worse calibration, with no one method consistently outperforming the others. \nThe authors show that thresholding uncertainty scores can lead to better correctness, safety, and less hallucination on unanswerable questions." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- **Limited novelty:** The novelty of the paper is pretty limited. From the abstract/intro, it seems like the main contribution of the paper is in showing that abstention based on uncertainty can improve results. This result is not new (see next point about missing related work). Moreover, the primary methodological novelty in this work is In-Dialogue Uncertainty, which is a fairly small contribution and does not consistently provide benefits in all settings. The Discussion presents a more nuanced view of the contribution (i.e. framing this paper as a survey of confidence estimation methods and showing that there isn't one method that consistently does well.) This framing would have been more novel but then I would have expected to see more different uncertainty estimation methods tested. \n- **Missing related work:** This paper misses a large chunk of the related work on abstention and confidence estimation from the last 2 years, focusing on older work. Examples:\n\t- https://arxiv.org/pdf/2407.18418\n\t- https://arxiv.org/abs/2308.13387\n\t- https://arxiv.org/abs/2311.09677\n\t- https://arxiv.org/abs/2404.00474\n\t- https://arxiv.org/abs/2405.21028\n\t- https://arxiv.org/abs/2401.06730\n\t- https://aclanthology.org/2024.naacl-long.301/\n\n- **Outdated models**: It's not clear why the authors only conduct experiments on Llama2, when there are many newer and more performant models available (even in the same family). To make a strong claim about when different estimation methods work and don't work, I would have expected to see more open-source models tested. \n- **No unified method**: one way this paper could have been made more compelling is if it presented a unified estimation method/recipe that worked well across settings. Currently, the paper does not have any such unified method." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 2 }, "desk_reject_comments": null, "details_of_ethics_concerns": { "value": "This paper utilizes human evaluations to conduct a safety assessment of the model's outputs. The authors state, \"We validate the effectiveness of fuzzy exact match by comparing it with human evaluations on 200 samples each from TriviaQA and SciQA.\" However, details regarding the background and diversity of these 200 individuals remain unclear, as well as whether these evaluations comply with IRB requirements." }, "flag_for_ethics_review": { "value": [ "Yes, Potentially harmful insights, methodologies and applications", "Yes, Responsible research practice (e.g., human subjects, data release)" ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "See above." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 2 }, "strengths": { "value": "1. The motivation behind this work is compelling and relevant for the deployment and real-world application of LLMs. However, while the authors highlight the benefits of combining RLHF with uncertainty to enhance performance, reviewers suggest that additional validation experiments, particularly in the areas of hallucination and safety, would strengthen the claims.\n2. The paper underscores the importance of uncertainty for abstention, demonstrating that incorporating uncertainty can improve various aspects of model performance." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The authors investigate the potential of uncertainty-based abstention to improve performance in question-answering tasks, specifically focusing on correctness, hallucinations, and safety scenarios. They analyze two types of uncertainty—statistical uncertainty and in-dialogue uncertainty—and examine the effects of RLHF on these uncertainties." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. Although the reviewers appreciate the study's motivation, they raise concerns regarding the experimental setup. For instance, in the hallucination settings, tests are only conducted on the SelfAware dataset. It would be beneficial to include additional datasets to more comprehensively evaluate the method's effectiveness in reducing hallucinations, especially given that current approaches primarily rely on Retrieval-Augmented Generation (RAG) [1].\n\n2. In the safety setting, the reviewers are interested in seeing how the uncertainty mechanism performs across a broader range of evaluation datasets. For example, PKU-SafeRLHF [3] provides safe, decoupled preferences and red-teaming prompts; how does the proposed approach perform on safety measures in these rigorous evaluations via case by case gpt-4 evaluation?\n\n3. The reviewers are not fully convinced by the claim that \"our experiments demonstrate that RLHF fine-tuning not only aligns the model with safety but also enhances its uncertainty awareness in relation to safety.\" RLHF alone does not guarantee model safety, particularly when the preference data distribution is uncertain. For instance, the GPT-4 technical report highlights that while RLHF helps align model responses with user intent, models may still exhibit brittle or undesired behaviors on both safe and unsafe inputs, especially when labeler instructions during reward model data collection are underspecified. Reviewers suggest that the authors provide a more detailed discussion on this aspect and include comparisons with models specifically designed for safety alignment, such as RLCD [2] and Safe RLHF [3].\n\n4. Regarding evaluation, the authors rely primarily on statistical measures, such as keyword-based approaches. However, this static evaluation method may fall short of detecting nuanced harmful responses, such as those involving emotional abuse. Additionally, Llama Guard’s performance drops in non-OOD (Out-of-Distribution) scenarios. Reviewers recommend including case-by-case GPT-4 evaluations to directly assess the safety of two responses, providing a more granular safety evaluation.\n\n[1] Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection \n[2] RLCD: Reinforcement Learning from Contrastive Distillation for Language Model Alignment \n[3] PKU-SafeRLHF: Towards Multi-Level Safety Alignment for LLMs with Human Preference" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "\"we recommend that practitioners differentiate between different types of questions before they decide whether to use statistical uncertainty or verbalized uncertainty for abstention...\" Could you explain why the experiment you conducted supports this claim?" }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "- Studied topic is timely and important. \n- Paper outlines multiple applications of the proposed method.\n- Paper is well-written and easy to follow. \n- Experiments test diverse uncertainty measures\n- Experiments consider different model sizes and variants." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper proposes uncertainty-based methods for abstaining from providing incorrect, hallucinated, or unsafe answers. It considers probabilistic uncertainty (log likelihood, entropy, and semantic uncertainty) as well as verbal uncertainty. Experiments with Llama2 models across various question-answering and adversarial-prompting benchmarks demonstrate that (1) the considered uncertainty measures contain information about whether an answer is incorrect, hallucinated, or unsafe, and (2) abstention based on these measures is effective." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "- There exist more sophisticated approaches to these problems but they are not compared empirically [1, 2]. [1] build classifiers based on hidden representations, and [2] constructs ensembles using LLM prompting. Authors mention the weaknesses of prompting and fine-tuning in related work but do not demonstrate them through experiments. In fact, the experiments do not seem to concern distribution shift so there is no reason not to compare with those methods. \n- Related work is missing some recent work showing similar results (e.g., [1,2,3]).\n- Experiment sections mostly discuss observations but does not attempt to explain the observed phenemena. \n- Some parts of the experiment section are unclear or can be further improved. Specifically, in figure 3, \"statistical uncertainty\" should be replaced with a specific measure and model (e.g., entropy). It is also missing model names. The plots need to have baseline curves to clearly illustrate improvements. \n\n[1] https://arxiv.org/abs/2304.13734\n\n[2] https://arxiv.org/abs/2402.00367\n\n[3] https://arxiv.org/abs/2402.13213" }, "withdrawal_confirmation": null }, { "TLDR": { "value": "Abstention based on the right form of uncertainty improves correctness, hallucinations and safety in LLMs." }, "_bibtex": { "value": "@inproceedings{\nanonymous2024uncertaintybased,\ntitle={Uncertainty-Based Abstention in {LLM}s Improves Safety and Reduces Hallucinations},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1DIdt2YOPw},\nnote={under review}\n}" }, "abstract": { "value": "A major barrier to the practical deployment of large language models (LLMs) is their lack of reliability. Three situations where this is particularly apparent are correctness, hallucinations when given unanswerable questions, and safety where responses are harmful or offensive. In all three cases, models should ideally abstain from responding---much like humans refrain from answering questions when uncertain. Inspired by analogous approaches in classification, this study explores the feasibility and efficacy of LLMs abstaining when uncertain in the domain of question-answering. We investigate two kinds of uncertainties, statistical uncertainty metrics and a distinct verbalized measure, termed as In Dialogue Uncertainty (InDU), measuring hedge words such as `I don't know' in responses. Using these uncertainty measures combined with models with and without reinforcement learning with human feedback (RLHF), we show in all three situations, abstention based on the right kind of uncertainty measure can boost the reliability of LLMs. By abstaining for a few highly uncertain samples we improve correctness by up to 8\\%, avoid 50\\% of hallucinations by correctly identifying unanswerable questions, and in particular increase safety by 70-99\\% with almost no additional computational overhead." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "LLMs", "uncertainty", "abstention", "correctness", "hallucinations", "safety" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/66216b5fd69ab306b3caa73bd9d7d3a92e0c88b6.pdf" }, "presentation": null, "primary_area": { "value": "alignment, fairness, safety, privacy, and societal considerations" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]

[ { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "as above" }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The paper present theories to show that DRO can help address the distribution matching problem in training and testing diffusion models.\n\n2. The improvement over baselines on Cifar and Imagenet64 show that DRO is useful." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "The paper proposes to introduce DRO to address the distribution matching problem at training diffusion model." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. There is no qualitative comparisons. Authors mainly conduct experiments on Cifar, ImageNet and Laion dataset. It would be better to put some images for more direct comparisons. In addition, the code is not provided. \n\n2. The efficiency comparison. I am wondering how much overhead it brings to adopt eq 14 instead of the classical denoising objective. I am expecting that it is quite large.\n\nI am giving score of 6 based on the prerequisite that above two concerns are answered during rebuttal." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 3 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 3 }, "primary_area": null, "questions": { "value": "1. some ablation studies for different perturbation levels $\\alpha$ should be given.\n2. Some discussions about different perturbation methods ($\\ell_1$, $\\ell_2$, or $\\ell_\\infty$) should be discussed." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. This paper performs a theoretical analysis of diffusion models and identifies the distribution mismatch problem.\n\n2. This paper further builds a connection between the distribution robust optimization and adversarial learning for diffusion models, and develops an adversarial training method for diffusion models.\n\n3. This paper conducts efficient adversarial training methods on both diffusion models and consistency models in many tasks. Experimental results demonstrate the effectiveness of the developed algorithms." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper studies the training of unconditional diffusion model. In particular, in order to achieve a better generation quality and enable robust learning of the score network, this paper develops a DRO-based method, and prove the DRO objective in training diffusion models can be formulated as an adversarial learning problem. The paper also identifies a similar mismatch issue in the recently proposed consistency model (CM) and demonstrates that AT can address this problem as well. The authors propose efficient AT for both DPM and CM, with empirical studies confirming the effectiveness of AT in enhancing diffusion-based models." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. In general, the algorithm developed in this paper is motivated by the distribution mismatch along the diffusion path. However, there is no experimental results to justify the motivation, there are also no experimental results to verify that the DRO framework can indeed help mitigate the distribution mismatch problem. \n\n2. Proposition 2 has already been discovered in existing theoretical papers [1], see their section 3.1. The authors should comment on this point around Proposition 2.\n\n3. The advantage of ADM-AT is not that significant compared with the ADM method, a more detailed ablation study or theoretical analysis on using adversarial noise or random Gaussian noise should be added.\n\n4. Some statements are not clearly presented. For instance, the description of ADM is not given, the norm notations $\\|\\|$ are abused, should that be $\\ell_1$, $\\ell_2$, or $\\ell_\\infty$?\n \n\n[1] Chen, Lee, and Lu, Improved Analysis of Score-based Generative Modeling: User-Friendly Bounds under Minimal Smoothness Assumptions, ICML 2023" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 4 }, "primary_area": null, "questions": { "value": "The paper is a good one in general. I like how the problem is formulated and how the solution is derived. However, given the current evaluation (see the weakness), I am not fully convinced the proposed method is an effective way to deal with the problem. I would like to see how the authors respond to my concerns." }, "rating": { "value": 5 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. Identifying and formulating the distribution mismatch problem in diffusion model is an important problem in practice.\n2. The proposed solution is elegant, supported by sufficient theoretical analysis. The derivations of the solution is clear and sound.\n3. The writing is fairly clear." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper identifies the distribution mismatch problem in the training and sampling processes. Consequently, they propose a distributionally robust optimization procedure in the training to bridge the gap. The authors apply the method to both diffusion models and the consistent model, and demonstrate the effectiveness of the proposed method on several benchmarks." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "My main concern on this paper is the evaluation. Currently the proposed method is only evaluated using the ADM model. I wonder whether the effectiveness on more advanced model such as the stable diffusion still holds?\n\nFurthermore, the authors only use FID score as the evaluation metric, while it is easy to evaluate the results using other metrics such as IS, sFID, precision, recall, as done in the ADM paper. Why these metrics are not included?" }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": null, "abstract": null, "anonymous_url": null, "authorids": null, "authors": null, "code_of_conduct": { "value": "Yes" }, "code_of_ethics": null, "comment": null, "confidence": { "value": 4 }, "contribution": { "value": 3 }, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": { "value": [ "No ethics review needed." ] }, "keywords": null, "large_language_models": null, "no_acknowledgement_section": null, "other_comments_on_LLMs": null, "paperhash": null, "pdf": null, "presentation": { "value": 2 }, "primary_area": null, "questions": { "value": "1. As the weakness above, for Table 1 and Table 2, more NFEs should also be verified.\n\n2. Why not also try generation using consistency models on benchmark datasets such as CIFAR10 and ImageNet, which can be more common and convincing?\n\n3. Derivations in supplementary material should be checked carefully and written with more details. \n\n4. Why efficient AT can improve performance compared with PGD is a bit confusing. Intuitively, PGD should be more accurate to find $\\delta_t$, thus more deep insights should be provided here." }, "rating": { "value": 6 }, "reciprocal_reviewing": null, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": { "value": 3 }, "strengths": { "value": "1. The motivation for mitigating distribution mismatching is clear and important for efficient sampling. \n\n2. This paper provides strong theoretical support for implementing adversarial training to correct distribution mismatching, making this method convincing." }, "student_author": null, "submission_guidelines": null, "summary": { "value": "This paper points out the distribution mismatching problem in traditional training of diffusion-based models (DPM) and proposes to conduct efficient adversarial training (AT) during the training of DPM to mitigate this problem. Theoretical analysis is strong enough to support its argument and experiments also verify the effectiveness of the proposed method." }, "supplementary_material": null, "title": null, "venue": null, "venueid": null, "weaknesses": { "value": "1. The experimental results may not be enough, for example, for Table 1 and Table 2, more NFEs should also be verified, although this method can improve efficient sampling, whether is adaptable and robust for more denoising steps should also be verified. \n\n2. Some complex derivations in supplementary material are too brief to understand, such as Eq(30) and Eq(59-62), I'm not sure if there are any typos in them, I suggest checking the equations carefully and modifying them." }, "withdrawal_confirmation": null }, { "TLDR": null, "_bibtex": { "value": "@inproceedings{\nanonymous2024improved,\ntitle={Improved Diffusion-based Generative Model with Better Adversarial Robustness},\nauthor={Anonymous},\nbooktitle={Submitted to The Thirteenth International Conference on Learning Representations},\nyear={2024},\nurl={https://openreview.net/forum?id=1DVgysiIt7},\nnote={under review}\n}" }, "abstract": { "value": "Diffusion Probabilistic Models (DPMs) have achieved considerable success in generation. However, its training and sampling processes are confronted with the problem of distribution mismatch. During the denoising process, the input data distributions of the model are different during the training and inference stages, which makes the model potentially generate inaccurate data. To obviate this, we conduct an analysis of the training objective of DPM, and theoretically prove that the mismatch can be mitigated by Distributionally Robust Optimization (DRO), which is equivalent to conducting robustness-driven Adversarial Training (AT) on the DPM. Furthermore, for the recently proposed consistency model (CM), which distills the inference process of the DPM, we prove that its training objective similarly faces the mismatch issue. Fortunately, such a problem is also mitigated by AT. Thereafter, we propose to conduct efficient AT on both DPM and CM. Finally, a series of empirical studies verify the effectiveness of AT in diffusion-based models." }, "anonymous_url": { "value": "I certify that there is no URL (e.g., github page) that could be used to find authors’ identity." }, "authorids": null, "authors": null, "code_of_conduct": null, "code_of_ethics": { "value": "I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics." }, "comment": null, "confidence": null, "contribution": null, "desk_reject_comments": null, "details_of_ethics_concerns": null, "flag_for_ethics_review": null, "keywords": { "value": [ "Generative Model; Adversarial Robustness; Diffusion Model; Distributional Robustness Optimization" ] }, "large_language_models": null, "no_acknowledgement_section": { "value": "I certify that there is no acknowledgement section in this submission for double blind review." }, "other_comments_on_LLMs": null, "paperhash": null, "pdf": { "value": "/pdf/35452c9c02243416f0b7374c1ed3cdb12e86bc86.pdf" }, "presentation": null, "primary_area": { "value": "generative models" }, "questions": null, "rating": null, "reciprocal_reviewing": { "value": "I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6." }, "resubmission": null, "revert_desk_rejection_confirmation": null, "revert_withdrawal_confirmation": null, "soundness": null, "strengths": null, "student_author": null, "submission_guidelines": { "value": "I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide." }, "summary": null, "supplementary_material": null, "title": { "value": "Improved Diffusion-based Generative Model with Better Adversarial Robustness" }, "venue": { "value": "ICLR 2025 Conference Submission" }, "venueid": { "value": "ICLR.cc/2025/Conference/Submission" }, "weaknesses": null, "withdrawal_confirmation": null } ]