abstract
stringlengths 383
2.62k
| TLDR
stringlengths 4
250
|
|---|---|
Federated Learning (FL) is a distributed machine learning paradigm that involves the cooperation of multiple clients to train a server model. In practice, it is hard to assume that each client possesses large-scale data or many clients are always available to participate in FL for the same round, which may lead to data deficiency. This deficiency degrades the entire learning process. To resolve this challenge, we propose a Federated learning with entropy-weighted ensemble Distillation and Self-supervised learning (FedDS). FedDS reliably deals with situations where not only the amount of data per client but also the number of clients is scarce. This advantage is achieved by leveraging the prevalent unlabeled data in the server. We demonstrate the effectiveness of FedDS on classification tasks for CIFAR-10/100 and PathMNIST. In CIFAR-10, our method shows the improvement over FedAVG by 12.54% in data deficient regime, and by 17.16% and 23.56% in more challenging scenarios of noisy label or Byzantine client cases, respectively. | A federated learning algorithm that tackles data deficiency by exploiting unlabeled data at server. |
Many recent breakthroughs of deep reinforcement learning (RL) are mainly built upon large-scale distributed training of model-free methods using millions to billions of samples. On the other hand, state-of-the-art model-based RL methods can achieve human-level sample efficiency but often take a much longer over all training time than model-free methods. However, high sample efficiency and fast training time are both important to many real-world applications. We develop SpeedyZero, a distributed RL system built upon a state-of-the-art model-based RL method, EfficientZero, with a dedicated system design for fast distributed computation. We also develop two novel algorithmic techniques, Priority Refresh and Clipped LARS, to stabilize training with massive parallelization and large batch size. SpeedyZero maintains on-par sample efficiency compared with EfficientZero while achieving a 14.5X speedup in wall-clock time, leading to human-level performances on the Atari benchmark within 35 minutes using only 300k samples. In addition, we also present an in-depth analysis on the fundamental challenges in further scaling our system to bring insights to the community. | SpeedyZero is a distributed model-based RL training system based on EfficientZero, featuring fast training speed and high sample efficiency. |
Sharpness-aware minimization (SAM) and related adversarial deep-learning methods can drastically improve generalization, but their underlying mechanisms are not yet fully understood. Here, we establish SAM as a relaxation of the Bayes objective where the expected negative-loss is replaced by the optimal convex lower bound, obtained by using the so-called Fenchel biconjugate. The connection enables a new Adam-like extension of SAM to automatically obtain reasonable uncertainty estimates, while sometimes also improving its accuracy. By connecting adversarial and Bayesian methods, our work opens a new path to robustness. | We show that SAM can be seen as a relaxation of Bayes, by using Fenchel conjugates. |
Large language models show improved downstream task performance when prompted to generate step-by-step reasoning to justify their final answers. These reasoning steps greatly improve model interpretability and verification, but objectively studying their correctness (independent of the final answer) is difficult without reliable methods for automatic evaluation. We simply do not know how often the stated reasoning steps actually support the final end task predictions. In this work, we present ROSCOE, a suite of interpretable, unsupervised automatic scores that improve and extend previous text generation evaluation metrics. To evaluate ROSCOE against baseline metrics, we design a typology of reasoning errors and collect synthetic and human evaluation scores on commonly used reasoning datasets. In contrast with existing metrics, ROSCOE can measure semantic consistency, logicality, informativeness, fluency, and factuality — among other traits — by leveraging properties of step-by-step rationales. We empirically verify the strength of our metrics on five human annotated and six programmatically perturbed diagnostics datasets - covering a diverse set of tasks that require reasoning skills and show that ROSCOE can consistently outperform baseline metrics. | We propose a new taxonomy for reasoning errors and suite of metrics to score step-by-step reasoning in language models. |
Algorithmic recourse is rising as a prominent technique to promote the explainability and transparency of the predictive model in ethical machine learning. Existing approaches to algorithmic recourse often assume an invariant predictive model; however, this model, in reality, is usually updated temporally upon the input of new data. Thus, a recourse that is valid respective to the present model may become invalid for the future model. To resolve this issue, we propose a pipeline to generate a model-agnostic recourse that is robust to model shifts. Our pipeline first estimates a linear surrogate of the nonlinear (black-box) model using covariance-robust minimax probability machines (MPM); then, the recourse is generated with respect to this robust linear surrogate. We show that the covariance-robust MPM recovers popular regularization schemes, including l_2-regularization and class-reweighting. We also show that our covariance-robust MPM pushes the decision boundary in an intuitive manner, which facilitates an interpretable generation of a robust recourse. The numerical results demonstrate the usefulness and robustness of our pipeline. | We propose a novel pipeline to generate a model-agnostic recourse that is robust to model shifts. |
Neural Representations have recently been shown to effectively reconstruct a wide range of signals from 3D meshes and shapes to images and videos. We show that, when adapted correctly, neural representations can be used to directly represent the weights of a pre-trained convolutional neural network, resulting in a Neural Representation for Neural Networks (NeRN). Inspired by coordinate inputs of previous neural representation methods, we assign a coordinate to each convolutional kernel in our network based on its position in the architecture, and optimize a predictor network to map coordinates to their corresponding weights. Similarly to the spatial smoothness of visual scenes, we show that incorporating a smoothness constraint over the original network's weights aids NeRN towards a better reconstruction. In addition, since slight perturbations in pre-trained model weights can result in a considerable accuracy loss, we employ techniques from the field of knowledge distillation to stabilize the learning process. We demonstrate the effectiveness of NeRN in reconstructing widely used architectures on CIFAR-10, CIFAR-100, and ImageNet. Finally, we present two applications using NeRN, demonstrating the capabilities of the learned representations. | In this paper we present NerN: a neural representation for the weights of a pretrained neural network, which is obtained by applying smoothness over the reconstructed weights and various knowledge distillation techniques |
Despite the clear performance benefits of data augmentations, little is known about why they are so effective. In this paper, we disentangle several key mechanisms through which data augmentations operate. Establishing an exchange rate between augmented and additional real data, we find that in out-of-distribution testing scenarios, augmentations which yield samples that are diverse, but inconsistent with the data distribution can be even more valuable than additional training data. Moreover, we find that data augmentations which encourage invariances can be more valuable than invariance alone, especially on small and medium sized training sets. Following this observation, we show that augmentations induce additional stochasticity during training, effectively flattening the loss landscape. | We uncover mechanisms by which data augmentations regularize training and inform the relationship between augmentations and extra data, invariance, stochasticity, and flatness. |
Removing background noise from speech audio has been the subject of considerable effort, especially in recent years due to the rise of virtual communication and amateur recordings. Yet background noise is not the only unpleasant disturbance that can prevent intelligibility: reverb, clipping, codec artifacts, problematic equalization, limited bandwidth, or inconsistent loudness are equally disturbing and ubiquitous. In this work, we propose to consider the task of speech enhancement as a holistic endeavor, and present a universal speech enhancement system that tackles 55 different distortions at the same time. Our approach consists of a generative model that employs score-based diffusion, together with a multi-resolution conditioning network that performs enhancement with mixture density networks. We show that this approach significantly outperforms the state of the art in a subjective test performed by expert listeners. We also show that it achieves competitive objective scores with just 4-8 diffusion steps, despite not considering any particular strategy for fast sampling. We hope that both our methodology and technical contributions encourage researchers and practitioners to adopt a universal approach to speech enhancement, possibly framing it as a generative task. | We propose to consider the task of speech enhancement as a universal endeavor, and provide a diffusion-based approach to deal with 55 different distortions at the same time. |
Unsupervised domain adaptation for video recognition is challenging where the domain shift includes both spatial variations and temporal dynamics. Previous works have focused on exploring contrastive learning for cross-domain alignment. However, limited variations in intra-domain positives, false cross-domain positives, and false negatives hinder contrastive learning from fulfilling intra-domain discrimination and cross-domain closeness. This paper presents a non-contrastive learning framework without relying on negative samples for unsupervised video domain adaptation. To address the limited variations in intra-domain positives, we set unlabeled target videos as anchors and explored to mine "informative intra-domain positives" in the form of spatial/temporal augmentations and target nearest neighbors (NNs).
To tackle the false cross-domain positives led by noisy pseudo-labels, we reversely set source videos as anchors and sample the synthesized target videos as "robust cross-domain positives" from an estimated target distribution, which are naturally more robust to the pseudo-label noise. Our approach is demonstrated to be superior to state-of-the-art methods through extensive experiments on several cross-domain action recognition benchmarks.
| We introduce the bottlenecks of in existing contrastive-based video DA methods and propose a unified solution to address them without relying on negatives by mining informative and robust intra-domain positives and cross-domain positives. |
State of the art results in reinforcement learning suggest that multi-step learning is necessary. However, the increased variance that comes with it makes it difficult to increase the update horizon beyond relatively small numbers. In this paper, we report the counterintuitive finding that decreasing the batch size substantially improves performance across a large swath of deep RL agents. It is well-known that gradient variance decreases with increasing batch sizes, so obtaining improved performance by increasing variance on two fronts is a rather surprising finding. We conduct a broad set of experiments to better understand this variance double-down phenomenon. | We perform an exhaustive investigation into the interplay of batch size and update horizon and uncover a surprising phenomenon: when increasing the update horizon, it is more beneficial to decrease the batch size |
The prediction of molecular properties is a crucial task in the field of material and drug discovery. The potential benefits of using deep learning techniques are reflected in the wealth of recent literature. Still, these techniques are faced with a common challenge in practice: Labeled data are limited by the cost of manual extraction from literature and laborious experimentation. In this work, we propose a data-efficient property predictor by utilizing a learnable hierarchical molecular grammar that can generate molecules from grammar production rules. Such a grammar induces an explicit geometry of the space of molecular graphs, which provides an informative prior on molecular structural similarity. The property prediction is performed using graph neural diffusion over the grammar-induced geometry. On both small and large datasets, our evaluation shows that this approach outperforms a wide spectrum of baselines, including supervised and pre-trained graph neural networks. We include a detailed ablation study and further analysis of our solution, showing its effectiveness in cases with extremely limited data (only ${\sim}100$ samples), and its extension to application in molecular generation. | We propose a data-efficient molecular property predictor based on an explicit geometry of the space of molecular graphs induced by a learnable hierarchical molecular grammar. |
A powerful framework for studying graphs is to consider them as geometric graphs: nodes are randomly sampled from an underlying metric space, and any pair of nodes is connected if their distance is less than a specified neighborhood radius. Currently, the literature mostly focuses on uniform sampling and constant neighborhood radius. However, real-world graphs are likely to be better represented by a model in which the sampling density and the neighborhood radius can both vary over the latent space. For instance, in a social network communities can be modeled as densely sampled areas, and hubs as nodes with larger neighborhood radius. In this work, we first perform a rigorous mathematical analysis of this (more general) class of models, including derivations of the resulting graph shift operators. The key insight is that graph shift operators should be corrected in order to avoid potential distortions introduced by the non-uniform sampling. Then, we develop methods to estimate the unknown sampling density in a self-supervised fashion. Finally, we present exemplary applications in which the learnt density is used to 1) correct the graph shift operator and improve performance on a variety of tasks, 2) improve pooling, and 3) extract knowledge from networks. Our experimental findings support our theory and provide strong evidence for our model. | We introduce geometric graphs with hubs, an effective model for real-world graphs, and retrieve the sampling density by which those graphs are sampled from continuous latent spaces, to achieve various tasks. |
Predicting the pose of objects from a single image is an important but difficult computer vision problem. Methods that predict a single point estimate do not predict the pose of objects with symmetries well and cannot represent uncertainty. Alternatively, some works predict a distribution over orientations in $\mathrm{SO}(3)$. However, training such models can be computation- and sample-inefficient. Instead, we propose a novel mapping of features from the image domain to the 3D rotation manifold. Our method then leverages $\mathrm{SO}(3)$ equivariant layers, which are more sample efficient, and outputs a distribution over rotations that can be sampled at arbitrary resolution. We demonstrate the effectiveness of our method at object orientation prediction, and achieve state-of-the-art performance on the popular PASCAL3D+ dataset. Moreover, we show that our method can model complex object symmetries, without any modifications to the parameters or loss function. Code is available at \url{https://dmklee.github.io/image2sphere}. | We propose a novel architecture which efficiently describes uncertainty in pose estimation from images by using learned SO(3)-equivariant features to generate complex distributions over SO(3) with the Fourier basis. |
Unsupervised learning plays an important role in many fields, such as machine learning, data compression, and neuroscience. Compared to static data, methods for extracting low-dimensional structure for dynamic data are lagging. We developed a novel information-theoretic framework, Compressed Predictive Information Coding (CPIC), to extract predictive latent representations from dynamic data. Predictive information quantifies the ability to predict the future of a time series from its past. CPIC selectively projects the past (input) into a low dimensional space that is predictive about the compressed data projected from the future (output). The key insight of our framework is to learn representations by balancing the minimization of compression complexity with maximization of the predictive information in the latent space. We derive tractable variational bounds of the CPIC loss by leveraging bounds on mutual information. The CPIC loss induces the latent space to capture information that is maximally predictive of the future of the data from the past. We demonstrate that introducing stochasticity in the encoder and maximizing the predictive information in latent space contributes to learning more robust latent representations. Furthermore, our variational approaches perform better in mutual information estimation compared with estimates under the Gaussian assumption commonly used. We show numerically in synthetic data that CPIC can recover dynamical systems embedded in noisy observation data with low signal-to-noise ratio. Finally, we demonstrate that CPIC extracts features more predictive of forecasting exogenous variables as well as auto-forecasting in various real datasets compared with other state-of-the-art representation learning models. Together, these results indicate that CPIC will be broadly useful for extracting low-dimensional dynamic structure from high-dimensional, noisy time-series data. | This work proposes a novel information-theoretic framework, Compressed Predictive Information Coding (CPIC), to extract predictive latent representations from dynamic data |
Fair classification aims to stress the classification models to achieve the equality (treatment or prediction quality) among different sensitive groups. However, fair classification can be under the risk of poisoning attacks which deliberately insert malicious training samples to manipulate the trained classifiers' performance. In this work, we study the poisoning scenario where the attacker can insert a small fraction of samples into training data, with arbitrary sensitive attributes as well as other predictive features. We demonstrate that the fairly trained classifiers can be greatly vulnerable to such poisoning attacks, with much worse accuracy & fairness trade-off, even when we apply some of the most effective defenses (originally proposed to defend traditional classification tasks). As countermeasures to defend fair classification tasks, we propose a general and theoretically guaranteed framework which accommodates traditional defense methods to fair classification against poisoning attacks. Through extensive experiments, the results validate that the proposed defense framework obtains better robustness in terms of accuracy and fairness than baseline methods. | We propose new poisoning attack and defense for fair classification methods. |
Language models are defined over a finite set of inputs, which creates a vocabulary bottleneck when we attempt to scale the number of supported languages. Tackling this bottleneck results in a trade-off between what can be represented in the embedding matrix and computational issues in the output layer. This paper introduces PIXEL, the Pixel-based Encoder of Language, which suffers from neither of these issues. PIXEL is a pretrained language model that renders text as images, making it possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. PIXEL is trained to reconstruct the pixels of masked patches instead of predicting a distribution over tokens. We pretrain the 86M parameter PIXEL model on the same English data as BERT and evaluate on syntactic and semantic tasks in typologically diverse languages, including various non-Latin scripts. We find that PIXEL substantially outperforms BERT on syntactic and semantic processing tasks on scripts that are not found in the pretraining data, but PIXEL is slightly weaker than BERT when working with Latin scripts. Furthermore, we find that PIXEL is more robust than BERT to orthographic attacks and linguistic code-switching, further confirming the benefits of modelling language with pixels. | We train PIXEL, a language model that operates solely on images of rendered text, and show that it is possible to transfer representations across languages based on orthographic similarity or the co-activation of pixels. |
Traditional stochastic sampling methods for open-ended neural text generation focus on truncating the low-likelihood part of the predicted distribution. They do not directly manipulate the high-likelihood part, which leads to the likelihood trap that induces repetition and boredom. They also do not directly leverage that human does not always favor high-likelihood texts. Inspired by these, we propose a novel sampling method that rescales the high-likelihood part of the distribution with inverse probability weighting. It increases the diversity by rescaling and penalizing the high-likelihood words, and preserves the fluency by using multi-filtering truncation on the low-likelihood words. We use pre-trained language models to compare our algorithm with traditional sampling methods. Results show that our algorithm can significantly increase the diversity and novelty of generated texts without corrupting the fluency. | A novel sampling algorithm for neural text generation with improved diversity and novelty compared with top-p/k and temperature sampling. |
Representation learning often plays a critical role in avoiding the curse of dimensionality in reinforcement learning. A representative class of algorithms exploits spectral decomposition of the stochastic transition dynamics to construct representations that enjoy strong theoretical properties in idealized settings. However, current spectral methods suffer from limited applicability because they are constructed for
state-only aggregation and are derived from a policy-dependent transition kernel, without considering the issue of exploration. To address these issues, we propose an alternative spectral method, Spectral Decomposition Representation (SPEDER), that extracts a state-action abstraction from the dynamics without inducing spurious dependence on the data collection policy, while also balancing the exploration-versus-exploitation trade-off during learning. A theoretical analysis establishes the sample efficiency of the proposed algorithm in both the online and offline settings. In addition, an experimental investigation demonstrates superior performance over current state-of-the-art algorithms across several RL benchmarks. | We propose a new spectral representation learning method that gets rid of the policy dependency and can be easily applied in downstream tasks. |
Federated Learning (FL) is a well-established technique for privacy preserving
distributed training. Much attention has been given to various aspects of FL training. A growing number of applications that consume FL-trained models, however, increasingly operate under dynamically and unpredictably variable conditions, rendering a single model insufficient. We argue for training a global “family of models” cost efficiently in a federated fashion. Training them independently for different tradeoff points incurs ≈ O(k) cost for any k architectures of interest, however.
Straightforward applications of FL techniques to recent weight-shared training
approaches is either infeasible or prohibitively expensive. We propose SuperFed — an architectural framework that incurs O(1) cost to co-train a large family ofmodels in a federated fashion by leveraging weight-shared learning. We achieve an order of magnitude cost savings on both communication and computation by proposing two novel training mechanisms: (a) distribution of weight-shared models to federated clients, (b) central aggregation of arbitrarily overlapping weight-shared model parameters. The combination of these mechanisms is shown to reach an order of magnitude (9.43x) reduction in computation and communication cost for training a 5*10^18-sized family of models, compared to independently training as few as k = 9 DNNs without any accuracy loss. | Federated Training of K models in O(1) (amortized) communication and computation cost. |
In this paper we improve the zero-shot generalization ability of language models via Mixture-Of-Memory Augmentation (MoMA), a mechanism that retrieves augmentation documents from multiple information corpora (“external memories”), with the option to “plug in” new memory at inference time. We develop a joint learning mechanism that trains the augmentation component with latent labels derived from the end retrieval task, paired with hard negatives from the memory mixture. We instantiate the model in a zero-shot dense retrieval setting by augmenting a strong T5-based retriever with MoMA. Our model, MoMA-DR, obtains strong zero-shot retrieval accuracy on the eighteen tasks included in the standard BEIR benchmark. It outperforms other dense retrieval models of similar scales and achieves comparable accuracy with systems that seek generalization from increased scales in encoder models or vector indices. Our analysis illustrates the necessity of augmenting with mixture-of-memory for robust generalization, the benefits of joint learning, and how MoMA-DR utilizes the plug-in memory at inference time without changing its parameters. We plan to open source our code. | We explore the potential of augmenting lanuguage models with mixture-of-memory and plugging in new corpus during inference, which leads to their enhanced generalization ability on the zero-shot dense retrieval task. |
We consider a hybrid reinforcement learning setting (Hybrid RL), in which an agent has access to an offline dataset and the ability to collect experience via real-world online interaction. The framework mitigates the challenges that arise in both pure offline and online RL settings, allowing for the design of simple and highly effective algorithms, in both theory and practice. We demonstrate these advantages by adapting the classical Q learning/iteration algorithm to the hybrid setting, which we call Hybrid Q-Learning or Hy-Q. In our theoretical results, we prove that the algorithm is both computationally and statistically efficient whenever the offline dataset supports a high-quality policy and the environment has bounded bilinear rank. Notably, we require no assumptions on the coverage provided by the initial distribution, in contrast with guarantees for policy gradient/iteration methods. In our experimental results, we show that Hy-Q with neural network function approximation outperforms state-of-the-art online, offline, and hybrid RL baselines on challenging benchmarks, including Montezuma’s Revenge. | We propose a new hybrid RL framework with access to both offline dataset and online interaction, and design a hybrid RL algorithm that is statistically and computationally efficient. |
Classical supervised learning assumes a stable relation between inputs and outputs. However, this assumption is often invalid in real-world scenarios where the input-output relation in the data depends on some hidden contexts. We formulate a more general setting where the training data is sampled from multiple unobservable domains, while different domains may possess semantically distinct input-output maps. Training data exhibits inherent conflict in this setting, rendering vanilla empirical risk minimization problematic. We propose to tackle this problem by introducing an allocation function that learns to allocate conflicting data to different prediction models, resulting in an algorithm that we term LEAF. We draw an intriguing connection between our approach and a variant of the Expectation-Maximization algorithm. We provide theoretical justifications for LEAF on its identifiability, learnability, and generalization error. Empirical results demonstrate the efficacy and potential applications of LEAF in a range of regression and classification tasks on both synthetic data and real-world datasets. | We formulate the problem of learning from conflicting data with hidden contexts and propose a subjective learning framework to tackle this problem. |
We propose a canonical approach for feature selection, sparse learnable masks (SLM). SLM integrates learnable sparse masks into end-to-end training. For the fundamental non-differentiability challenge of selecting a desired number of features, we propose duo mechanisms for automatic mask scaling to achieve the desired feature sparsity, and gradually tempering this sparsity for effective learning. In addition, SLM employs a novel objective that maximizes the mutual information between the selected features and the labels. Empirically, SLM achieves state-of-the-art results on several benchmark datasets, often by a significant margin, especially on real-world challenging datasets. | SLM is an end-to-end feature selection method using a sparse learnable mask and a novel mutual information maximizer. |
The locally balanced informed proposal has proved to be highly effective for sampling from discrete spaces. However, its success relies on the "local'' factor, which ensures that whenever the proposal distribution is restricted to be near the current state, the locally balanced weight functions are asymptotically optimal and the gradient approximations are accurate. In seeking a more efficient sampling algorithm, many recent works have considered increasing the scale of the proposal distributions, but this causes the "local'' factor to no longer hold. Instead, we propose any-scale balanced samplers to repair the gap in non-local proposals. In particular, we substitute the locally balanced function with an any-scale balanced function that can self-adjust to achieve better efficiency for proposal distributions at any scale. We also use quadratic approximations to capture curvature of the target distribution and reduce the error in the gradient approximation, while employing a Gaussian integral trick with a special estimated diagonal to efficiently sample from the quadratic proposal distribution. On various synthetic and real distributions, the proposed sampler substantially outperforms existing approaches. | We identify two key issues of existing gradient based locally balanced samplers, and provide improved proposals with adjusted weight function and 2nd order approximation. |
Behavior constrained policy optimization has been demonstrated to be a successful paradigm for tackling Offline Reinforcement Learning. By exploiting historical transitions, a policy is trained to maximize a learned value function while constrained by the behavior policy to avoid a significant distributional shift. In this paper, we propose our closed-form policy improvement operators. We make a novel observation that the behavior constraint naturally motivates the use of first-order Taylor approximation, leading to a linear approximation of the policy objective. Additionally, as practical datasets are usually collected by heterogeneous policies, we model the behavior policies as a Gaussian Mixture and overcome the induced optimization difficulties by leveraging the LogSumExp's lower bound and Jensen's Inequality, giving rise to a closed-form policy improvement operator. We instantiate an offline RL algorithm with our novel policy improvement operator and empirically demonstrate its effectiveness over state-of-the-art algorithms on the standard D4RL benchmark. | We proposed a closed-form policy improvement operator and modeled the behavior policies as a Gaussian Mixture. |
Diffusion-based generative models learn to iteratively transfer unstructured noise to a complex target distribution as opposed to Generative Adversarial Networks (GANs) or the decoder of Variational Autoencoders (VAEs) which produce samples from the target distribution in a single step. Thus, in diffusion models every sample is naturally connected to a random trajectory which is a solution to a learned stochastic differential equation (SDE). Generative models are only concerned with the final state of this trajectory that delivers samples from the desired distribution. \cite{abstreiter2021diffusion} showed that these stochastic trajectories can be seen as continuous filters that wash out information along the way. Consequently, it is reasonable to ask if there is an intermediate time step at which the preserved information is optimal for a given downstream task. In this work, we show that a combination of information content from different time steps gives a strictly better representation for the downstream task. We introduce an attention and recurrence based modules that ``learn to mix'' information content of various time-steps such that the resultant representation leads to superior performance in downstream tasks.
| We perform an analysis on the trajectory-based representation obtained from Diffusion Based Representation Learning to measure how different points of the trajectory encode semantically different information. |
Complex nonlinear interplays of multiple scales give rise to many interesting physical phenomena and pose major difficulties for the computer simulation of multiscale PDE models in areas such as reservoir simulation, high frequency scattering and turbulence modeling. In this paper, we introduce a hierarchical transformer (HT-Net) scheme to efficiently learn the solution operator for multiscale PDEs. We construct a hierarchical architecture with scale adaptive interaction range, such that the features can be computed in a nested manner and with a controllable linear cost. Self-attentions over a hierarchy of levels can be used to encode and decode the multiscale solution space over all scale ranges. In addition, we adopt an empirical $H^1$ loss function to counteract the spectral bias of the neural network approximation for multiscale functions. In the numerical experiments, we demonstrate the superior performance of the HT-Net scheme compared with state-of-the-art (SOTA) methods for representative multiscale problems.
| We design a hierarchical transformer based operator learning method, so that the accurate, efficient and robust computer simulation of multiscale PDE problems with an ensemble of input parameters becomes feasible. |
Offline Reinforcement Learning (ORL) enables us to separately study the two interlinked processes of reinforcement learning: collecting informative experience and inferring optimal behaviour. The second step has been widely studied in the offline setting, but just as critical to data-efficient RL is the collection of informative data. The task-agnostic setting for data collection, where the task is not known a priori, is of particular interest due to the possibility of collecting a single dataset and using it to solve several downstream tasks as they arise. We investigate this setting via curiosity-based intrinsic motivation, a family of exploration methods which encourage the agent to explore those states or transitions it has not yet learned to model. With Explore2Offline, we propose to evaluate the quality of collected data by transferring the collected data and inferring policies with reward relabelling and standard offline RL algorithms. We evaluate a wide variety of data collection strategies, including a new exploration agent, Intrinsic Model Predictive Control (IMPC), using this scheme and demonstrate their performance on various tasks. We use this decoupled framework to strengthen intuitions about exploration and the data prerequisites for effective offline RL. | We compare existing and new exploration methods as a new way to generate useful data for offline reinforcement learning. |
The growing popularity of machine learning models has led to their increased application in domains directly impacting human lives. In critical fields such as healthcare, banking, and criminal justice, tools that ensure trust and transparency are vital for the responsible adoption of these models. One such tool is \emph{actionable recourse} (AR) for negatively impacted users. AR describes recommendations of cost-efficient changes to a user's \emph{actionable} features to help them obtain favorable outcomes. Existing approaches for providing recourse optimize for properties such as proximity, sparsity, validity, and distance-based costs. However, an often-overlooked but crucial requirement for actionability is a consideration of \emph{User Preference} to guide the recourse generation process. Moreover, existing works considering a user's preferences require users to precisely specify their costs for taking actions. This requirement raises questions about the practicality of the corresponding solutions due to the high cognitive loads imposed. In this work, we attempt to capture user preferences via soft constraints in three simple forms: \textit{i) scoring continuous features, ii) bounding feature values} and \textit{iii) ranking categorical features}. We propose an optimization framework that is sensitive to {user preference} and a gradient-based approach to identify \emph{User Preferred Actionable Recourse (UP-AR)}. We empirically demonstrate the proposed approach's superiority in adhering to user preference while maintaining competitive performance in traditional metrics with extensive experiments. | Capturing user preference and suggesting actionable recourse for adversely affected individuals by a machine learning model. |
We present Amos, a stochastic gradient-based optimizer designed for training deep neural networks. It can be viewed as an Adam optimizer with theoretically supported, adaptive learning-rate decay and weight decay. A key insight behind Amos is that it leverages model-specific information to determine the initial learning-rate and decaying schedules. When used for pre-training BERT variants and T5, Amos consistently converges faster than the state-of-the-art settings of AdamW, achieving better validation loss within <=70% training steps and time, while requiring <=51% memory for slot variables. | An optimizer that consistently converges faster (<=70% training steps) than AdamW for pre-training Transformer variants. |
Federated learning (FL) is a popular distributed machine learning paradigm dealing with distributed and private data sets. Based on the data partition pattern, FL is often categorized into horizontal, vertical, and hybrid settings. All three settings have many applications, but the hybrid FL remains relatively less explored, because it deals with the challenging situation where both the feature space and the data samples are heterogeneous.
This work designs a novel mathematical model that effectively allows the clients to aggregate distributed data with heterogeneous, and possibly overlapping features and samples. Our main idea is to partition each client's model into a feature extractor part and a classifier part, where the former can be used to process the input data, while the latter is used to perform the learning from the extracted features. The heterogeneous feature aggregation is done through building a server model, which assimilates local classifiers and feature extractors through a carefully designed matching mechanism. A communication-efficient algorithm is then designed to train both the client and server models. Finally, we conducted numerical experiments on multiple image classification data sets to validate the performance of the proposed algorithm. To our knowledge, this is the first formulation and algorithm developed for hybrid FL. | In this paper, we proposed the first hybrid federated learning model and algorithm, which deals with partially overlapped features and samples in clients' datasets |
3D object detection from multiple image views is a fundamental and challenging task for visual scene understanding. Owing to its low cost and high efficiency, multi-view 3D object detection has demonstrated promising application prospects. However, accurately detecting objects through perspective views is extremely difficult due to the lack of depth information. Current approaches tend to adopt heavy backbones for image encoders, making them inapplicable for real-world deployment. Different from the images, LiDAR points are superior in providing spatial cues, resulting in highly precise localization. In this paper, we explore the incorporation of LiDAR-based detectors for multi-view 3D object detection. Instead of directly training a depth prediction network, we unify the image and LiDAR features in the Bird-Eye-View (BEV) space and adaptively transfer knowledge across non-homogenous representations in a teacher-student paradigm. To this end, we propose BEVDistill, a cross-modal BEV knowledge distillation (KD) framework for multi-view 3D object detection.
Extensive experiments demonstrate that the proposed method outperforms current KD approaches on a highly-competitive baseline, BEVFormer, without introducing any extra cost in the inference phase. Notably, our best model achieves 59.4 NDS on the nuScenes test leaderboard, achieving new state-of-the-arts in comparison with various image-based detectors. Code will be available at https://github.com/zehuichen123/BEVDistill. | We leverage LiDAR-based knowledge into multi-view 3d detectors with cross-modal BEV distillation. |
Data augmentations are effective in improving the invariance of learning machines. We argue that the core challenge of data augmentations lies in designing data transformations that preserve labels. This is relatively straightforward for images, but much more challenging for graphs. In this work, we propose GraphAug, a novel automated data augmentation method aiming at computing label-invariant augmentations for graph classification. Instead of using uniform transformations as in existing studies, GraphAug uses an automated augmentation model to avoid compromising critical label-related information of the graph, thereby producing label-invariant augmentations at most times. To ensure label-invariance, we develop a training method based on reinforcement learning to maximize an estimated label-invariance probability. Experiments show that GraphAug outperforms previous graph augmentation methods on various graph classification tasks. | We propose GraphAug, a novel automated data augmentation method aiming at computing label-invariant augmentations for graph classification. |
In the mode connectivity literature, it is widely accepted that there are common circumstances in which two neural networks, trained similarly on the same data, will maintain loss when interpolated in the weight space. In particular, transfer learning is presumed to ensure the necessary conditions for linear mode connectivity across training runs. In contrast to existing results from image classification, we find that among text classifiers (trained on MNLI, QQP, and CoLA), some pairs of finetuned models have large barriers of increasing loss on the linear paths between them. On each task, we find distinct clusters of models which are linearly connected on the test loss surface, but are disconnected from models outside the cluster---models that occupy separate basins on the surface. By measuring performance on specially-crafted diagnostic datasets, we find that these clusters correspond to different generalization strategies. For example, on MNLI, one cluster behaves like a bag of words model under domain shift, while another cluster uses syntactic heuristics. Our work demonstrates how the geometry of the loss surface can guide models towards different heuristic functions in standard finetuning settings. | Basins on the in-domain test loss surface predict generalization strategies for NLI, paraphrase, and CoLA tasks. |
As more practical and scalable quantum computers emerge, much attention has been focused on realizing quantum supremacy in machine learning. Existing quantum ML methods either (1) embed a classical model into a target Hamiltonian to enable quantum optimization or (2) represent a quantum model using variational quantum circuits and apply classical gradient-based optimization. The former method leverages the power of quantum optimization but only supports simple ML models, while the latter provides flexibility in model design but relies on gradient calculation, resulting in barren plateau (i.e., gradient vanishing) and frequent classical-quantum interactions. To address the limitations of existing quantum ML methods, we introduce Quark, a gradient-free quantum learning framework that optimizes quantum ML models using quantum optimization. Quark does not rely on gradient computation and therefore avoids barren plateau and frequent classical-quantum interactions. In addition, Quark can support more general ML models than prior quantum ML methods and achieves a dataset-size-independent optimization complexity. Theoretically, we prove that Quark can outperform classical gradient-based methods by reducing model query complexity for highly non-convex problems; empirically, evaluations on the Edge Detection and Tiny-MNIST tasks show that Quark can support complex ML models and significantly reduce the number of measurements needed for discovering near-optimal weights for these tasks. | A new quantum learning framework for classification task |
Physics-informed machine learning (PIML) aims to incorporate physics knowledge into deep neural networks (DNNs) to improve the model generalization. However, existing methods in PIML are either designed for specific problems or hard to interpret the results using black-box DNNs. In this work, we propose Taylor Neural Network (TaylorNet), a generic neural architecture that parameterizes Taylor polynomials using DNNs without using non-linear activation functions. The key challenges of developing TaylorNet lie in: (i) mitigating the curse of dimensionality caused by higher-order terms, and (ii) improving the stability of model training. To overcome these challenges, we first adopt Tucker decomposition to decompose the higher-order derivatives in Taylor expansion parameterized by DNNs into low-rank tensors. Then we propose a novel reducible TaylorNet to further reduce the computational complexity by removing more redundant parameters in the hidden layers. In order to improve training accuracy and stability, we develop a new Taylor initialization method. Finally, the proposed models are evaluated on a broad spectrum of applications, including image classification, natural language processing (NLP), and dynamical systems. The results demonstrate that our proposed Taylor-Mixer, which replaces MLP and activation layers in the MLP-Mixer with Taylor layer, can achieve comparable accuracy on image classification, and similarly on sentiment analysis in NLP, while significantly reducing the number of model parameters. More importantly, our method can interpret some dynamical systems with Taylor polynomials. Meanwhile, the results demonstrate that our Taylor initialization can significantly improve classification accuracy compared to Xavier and Kaiming initialization. | We propose a generic neural architecture, called TaylorNet, that can introduce inductive bias to DNNs with Taylor series expansion |
Gaussian processes (GPs) are powerful but computationally expensive machine learning models, requiring an estimate of the kernel covariance matrix for every prediction. In large and complex domains, such as graphs, sets, or images, the choice of suitable kernel can also be non-trivial to determine, providing an additional obstacle to the learning task. Over the last decade, these challenges have resulted in significant advances being made in terms of scalability and expressivity, exemplified by, e.g., the use of inducing points and neural network kernel approximations.
In this paper, we propose inducing Gaussian process networks (IGN), a simple framework for simultaneously learning the feature space as well as the inducing points. The inducing points, in particular, are learned directly in the feature space, enabling a seamless representation of complex structured domains while also facilitating scalable gradient-based learning methods.
We consider both regression and (binary) classification tasks and report on experimental results for real-world data sets showing that IGNs provide significant advances over state-of-the-art methods. We also demonstrate how IGNs can be used to effectively model complex domains using neural network architectures. | We introduce a new method to efficiently learn the kernel and inducing points for Gaussian processes. |
Adversarial training is a standard method to train neural networks to be robust to adversarial perturbation. However, in contrast with benign overfitting in the standard deep learning setting, which means that over-parameterized neural networks surprisingly generalize well for unseen data, while adversarial training method is able to achieve low robust training error, there still exists a significant robust generalization gap, which promotes us exploring what mechanism leads to robust overfitting during learning process. In this paper, we propose an implicit bias called $\textit{robust memorization}$ in adversarial training under the realistic data assumption. By function approximation theory, we prove that ReLU nets with efficient size have the ability to achieve robust memorization, while robust generalization requires exponentially large models. Then, we demonstrate robust memorization in adversarial training from both empirical and theoretical perspectives. In particular, we empirically investigate the dynamics of loss landscape over input, and we also provide theoretical analysis of robust memorization on data with linear separable assumption. Finally, we prove novel generalization bounds based on robust memorization, which further explains why deep neural networks have both high clean test accuracy and robust overfitting at the same time. | We provide a theoretical understanding of adversarial training by proposing a novel implicit bias called robust memorization. |
Existing dialogue modeling methods have achieved promising performance on various dialogue tasks with the aid of Transformer and the large-scale pre-trained language models. However, some recent studies revealed that the context representations produced by these methods suffer the problem of anisotropy. In this paper, we find that the generated representations are also not conversational, losing the conversation structure information during the context modeling stage. To this end, we identify two properties in dialogue modeling, i.e., locality and isotropy, and present a simple method for dialogue representation calibration, namely SimDRC, to build isotropic and conversational feature spaces. Experimental results show that our approach significantly outperforms current state-of-the-art models on three open-domain dialogue tasks with eight benchmarks. More in-depth analyses further confirm the effectiveness of our proposed approach. We release the code at https://github.com/hahahawu/SimDRC. | We present a simple dialogue representation calibration method to learn isotropic and conversational features during the dialogue modeling stage. |
Can we design a GNN that is accurate and interpretable at the same time? Could it also be robust to handle the case of homophily, heterophily, or even noisy edges without network effects? We propose SlenderGNN that has all desirable properties: (a) accurate, (b) robust, and (c) interpretable. For the reasons of its success, we had to dig deeper: The result is our GNNLIN framework which highlights the fundamental differences among popular GNN models (e.g., feature combination, structural normalization, etc.) and thus reveals the reasons for the success of our SlenderGNN, as well as the reasons for occasional failures of other GNN variants. Thanks to our careful design, SlenderGNN passes all the 'sanity checks' we propose, and it achieves the highest overall accuracy on 9 real-world datasets of both homophily and heterophily graphs, when compared against 10 recent GNN models. Specifically, SlenderGNN exceeds the accuracy of linear GNNs and matches or exceeds the accuracy of nonlinear models with up to 64 times fewer parameters. | We propose SlenderGNN, a linear GNN whose motivations are derived from comprehensive linearization on existing models. |
This work studies the explainability of graph neural networks (GNNs), which is important for the credibility of GNNs in practical usage. Existing work mostly follows the two-phase paradigm to interpret a prediction: feature attribution and selection. However, another important component --- regularization, which is crucial to facilitate the above paradigm --- has been seldom studied. In this work, we explore the role of regularization in GNNs explainability from the perspective of information theory. Our main findings are: 1) regularization is essentially pursuing the balance between two phases, 2) its optimal coefficient is proportional to the sparsity of explanations, 3) existing methods imply an implicit regularization effect of stochastic mechanism, and 4) its contradictory effects on two phases are responsible for the out-of-distribution (OOD) issue in post-hoc explainability. Based on these findings, we propose two common optimization methods, which can bolster the performance of the current explanation methods via sparsity-adaptive and OOD-resistant regularization schemes. Extensive empirical studies validate our findings and proposed methods. Code is available at https://anonymous.4open.science/r/Rethink_Reg-07F0.
| We rethink the role of regularization in GNNs explainability from the perspective of information theory, and propose four intriguing propositions of regularization. |
This paper studies model transferability when human decision subjects respond to a deployed machine learning model. In our setting, an agent or a user corresponds to a sample $(X,Y)$ drawn from a distribution $\mathcal{D}$ and will face a model $h$ and its classification result $h(X)$. Agents can modify $X$ to adapt to $h$, which will incur a distribution shift on $(X,Y)$. Therefore, when training $h$, the learner will need to consider the subsequently ``induced" distribution when the output model is deployed. Our formulation is motivated by applications where the deployed machine learning models interact with human agents, and will ultimately face \emph{responsive} and interactive data distributions. We formalize the discussions of the transferability of a model by studying how the model trained on the available source distribution (data) would translate to the performance on the induced domain. We provide both upper bounds for the performance gap due to the induced domain shift, as well as lower bound for the trade-offs that a classifier has to suffer on either the source training distribution or the induced target distribution. We provide further instantiated analysis for two popular domain adaptation settings with covariate shift and target shift. | This paper studies model transferability when human decision subjects respond to a deployed machine learning model. |
Offline reinforcement learning algorithms still lack trust in practice due to the risk that the learned policy performs worse than the original policy that generated the dataset or behaves in an unexpected way that is unfamiliar to the user. At the same time, offline RL algorithms are not able to tune their most important hyperparameter - the proximity of the learned policy to the original policy. We propose an algorithm that allows the user to tune this hyperparameter at runtime, thereby addressing both of the above mentioned issues simultaneously. This allows users to start with the original behavior and grant successively greater deviation, as well as stopping at any time when the policy deteriorates or the behavior is too far from the familiar one. | Offline RL policies need to be adaptvie after training so that a user can alter its behavior to its needs. |
Adaptive optimization methods are well known to achieve superior convergence relative to vanilla gradient methods. The traditional viewpoint in optimization, particularly in convex optimization, explains this improved performance by arguing that, unlike vanilla gradient schemes, adaptive algorithms mimic the behavior of a second-order method by adapting to the global geometry of the loss function. We argue that in the context of neural network optimization, this traditional viewpoint is insufficient. Instead, we advocate for a local trajectory analysis. For iterate trajectories produced by running a generic optimization algorithm OPT, we introduce $R^{\text{OPT}}_{\text{med}}$, a statistic that is analogous to the condition number of the loss Hessian evaluated at the iterates. Through extensive experiments, we show that adaptive methods such as Adam bias the trajectories towards regions where $R^{\text{Adam}}_{\text{med}}$ is small, where one might expect faster convergence. By contrast, vanilla gradient methods like SGD bias the trajectories towards regions where $R^{\text{SGD}}_{\text{med}}$ is comparatively large. We complement these empirical observations with a theoretical result that provably demonstrates this phenomenon in the simplified setting of a two-layer linear network. We view our findings as evidence for the need of a new explanation of the success of adaptive methods, one that is different than the conventional wisdom. | We study the difference between the local geometry of the training objective in deep learning using Adaptive algorithms and SGD. |
Machine learning algorithms are increasingly being deployed for high-stakes scenarios. A sizeable proportion of currently deployed models make their decisions in a black box manner. Such decision-making procedures are susceptible to intrinsic biases, which has led to a call for accountability in deployed decision systems. In this work, we focus on user-specified accountability of decision-making processes of black box systems. Previous work has formulated this problem as run time fairness monitoring over decision functions. However, formulating appropriate specifications for situation-appropriate fairness metrics is challenging. We construct AVOIR, an automated inference-based optimization system that improves bounds for and generalizes prior work across a wide range of fairness metrics. AVOIR offers an interactive and iterative process for exploring fairness violations aligned with governance and regulatory requirements. Our bounds improve over previous probabilistic guarantees for such fairness grammars in online settings. We also construct a novel visualization mechanism that can be used to investigate the context of reported fairness violations and guide users towards meaningful and compliant fairness specifications. We then conduct case studies with fairness metrics on three different datasets and demonstrate how the visualization and improved optimization can detect fairness violations more efficiently and ameliorate the issues with faulty fairness metric design. | A visual inference-based optimization framework that facilitates the specification and auditing of fairness on blackbox ML models efficiently. |
Group distributionally robust optimization, which aims to improve robust accuracies such as worst-group or unbiased accuracy, is one of the mainstream algorithms to mitigate spurious correlation and reduce dataset bias. While existing approaches have apparently gained performance in robust accuracy, these improvements mainly come from a trade-off at the expense of average accuracy. To address the challenges, we first propose a simple class-specific scaling strategy to control the trade-off between robust and average accuracies flexibly and efficiently, which is directly applicable to existing debiasing algorithms without additional training; it reveals that a naive ERM baseline matches or even outperforms the recent debiasing approaches by adopting the class-specific scaling. Then, we employ this technique to 1) evaluate the performance of existing algorithms in a comprehensive manner by introducing a novel unified metric that summarizes the trade-off between the two accuracies as a scalar value and 2) develop an instance-wise adaptive scaling technique for overcoming the trade-off and improving the performance even further in terms of both accuracies. Experimental results verify the effectiveness of the proposed frameworks in both tasks.
| We propose a simple class-specific scaling strategy to control the trade-off between robust and average accuracies, and based on this, we develop a comprehensive performance evaluation metric and advanced algorithm to improve the trade-off. |
One of the challenges in multivariate time series modeling is that changes in signals occur with different frequencies, even when the sampling rate is consistent across signals. In the case of multivariate time series prediction, the outcome is also determined by patterns of different frequencies. These encapsulate both long-term and short-term effects, which have so far not been sufficiently leveraged by deep learning time series models. We fill this gap by introducing a framework, called MultiWave, which augments any deep learning time series model with components operating at the intrinsic frequencies of the signals. MultiWave applies wavelet decomposition on each signal to obtain subsignals of different frequencies and groups all subsignals in the same frequency band together to train a component. The output of the components is combined through a gating mechanism that removes irrelevant frequencies for the given predictive task. We show that MultiWave accurately determines the informative frequency bands and that the augmented models including components trained to operate on those bands outperform the original models. We further show that applying MultiWave on top of different deep learning models improves their performance in several real-world applications. | In multivariate time-series datasets changes in signals occurs in different frequencies. MultiWave decomposes signals into different frequencies removes the irrelevant frequencies and models each group using a model component. |
Stein variational gradient descent (SVGD) \citep{DBLP:conf/nips/LiuW16} is a particle-based technique for Bayesian inference. SVGD has recently gained popularity because it combines the ability of variational inference to handle tall data with the modeling power of non-parametric inference. Unfortunately, the number of particles required to represent a model adequately grows exponentially with the dimensionality of the model. Stein mixtures \citep{nalisnick2017variational} alleviate the exponential growth in particles by letting each particle parameterize a distribution. However, the inference algorithm proposed by \cite{nalisnick2017variational} can be numerically unstable. We show that their algorithm corresponds to inference with the R\'enyi $\alpha$-divergence for $\alpha=0$ and that using other values for $\alpha$ can lead to more stable inference. We empirically study the performance of Stein mixtures inferred with different $\alpha$ values on various real-world problems, demonstrating significantly improved results when using $\alpha=1$, which coincides with using the evidence lower bound (ELBO). We call this instance of our algorithm ELBO-within-Stein. A black-box version of the inference algorithm (for arbitrary $\alpha\in \sR$) is available in the deep probabilistic programming language NumPyro \citep{phan2019}. | Stein mixture can be viewed as matching variational- to target-posterior by the Renyi divergence. This leads to a whole class of inference methods using the Renyi divergence's order. |
The training success, training speed and generalization ability of neural networks rely crucially on the choice of random parameter initialization. It has been shown for multiple architectures that initial dynamical isometry is particularly advantageous. Known initialization schemes for residual blocks, however, miss this property and suffer from degrading separability of different inputs for increasing depth and instability without Batch Normalization or lack feature diversity. We propose a random initialization scheme, Risotto, that achieves perfect dynamical isometry for residual networks with ReLU activation functions even for finite depth and width. It balances the contributions of the residual and skip branches unlike other schemes, which initially bias towards the skip connections. In experiments, we demonstrate that in most cases our approach outperforms initialization schemes proposed to make Batch Normalization obsolete, including Fixup and SkipInit, and facilitates stable training. Also in combination with Batch Normalization, we find that Risotto often achieves the overall best result. | We derive an initialization scheme for ResNets that induces perfect dynamical isometry at initialization. |
Synthetic datasets are often used to pretrain end-to-end optical flow networks, due to the lack of a large amount of labeled, real scene data. But major drops in accuracy occur when moving from synthetic to real scenes. How do we better transfer the knowledge learned from synthetic to real domains? To this end, we propose CLIP-Flow, a semi-supervised iterative pseudo labeling framework to transfer the pretraining knowledge to the target real domain. We leverage large-scale, unlabeled real data to facilitate transfer learning with the supervision of iteratively updated pseudo ground truth labels, bridging the domain gap between the synthetic and the real. In addition, we propose a contrastive flow loss on reference features and the warped features by pseudo ground truth flows, to further boost the accurate matching and dampen the mismatching due to motion, occlusion, or noisy pseudo labels. We adopt RAFT as the backbone and obtain an F1-all error of 4.11%, i.e., a 19% error reduction from RAFT (5.10%) and ranking 2nd place at submission on KITTI 2015 benchmark. Our framework can also be extended to other models, e.g., CRAFT, reducing the F1-all error from 4.79% to 4.66% on KITTI 2015 benchmark. | A semi-supervised framework for optical flow with iterative pseudo labeling and contrastive flow loss to facilitate representation learning with unlabeled data |
There is a growing interest in using spiking neural networks (SNNs) to study the brain \textit{in silico} and in emulating them on neuromorphic computers due to their lower energy consumption compared to artificial neural networks (ANNs). Significant progress has been made in directly training SNNs to perform on a par with ANNs in terms of accuracy. However, these methods are slow due to their sequential nature and require careful network regularisation to avoid overfitting. We propose a new SNN model, the $d$-block model, with stochastic absolute refractory periods and recurrent conductance latencies, which reduces the number of sequential computations using fast vectorised operations. Our model obtains accelerated training speeds and state-of-the-art performance across various neuromorphic datasets without the need for any regularisation and using fewer spikes compared to standard SNNs. | We propose a new SNN model which obtains accelerated training and state-of-the-art performance across various neuromorphic datasets without the need of any regularisation and using less spikes compared to standard SNNs. |
We develop and analyze DASHA: a new family of methods for nonconvex distributed optimization problems. When the local functions at the nodes have a finite-sum or an expectation form, our new methods, DASHA-PAGE, DASHA-MVR and DASHA-SYNC-MVR, improve the theoretical oracle and communication complexity of the previous state-of-the-art method MARINA by Gorbunov et al. (2020). In particular, to achieve an $\varepsilon$-stationary point, and considering the random sparsifier Rand$K$ as an example, our methods compute the optimal number of gradients $\mathcal{O}\left(\frac{\sqrt{m}}{\varepsilon\sqrt{n}}\right)$ and $\mathcal{O}\left(\frac{\sigma}{\varepsilon^{3/2}n}\right)$ in finite-sum and expectation form cases, respectively, while maintaining the SOTA communication complexity $\mathcal{O}\left(\frac{d}{\varepsilon \sqrt{n}}\right)$. Furthermore, unlike MARINA, the new methods DASHA, DASHA-PAGE and DASHA-MVR send compressed vectors only, which makes them more practical for federated learning. We extend our results to the case when the functions satisfy the Polyak-Lojasiewicz condition. Finally, our theory is corroborated in practice: we see a significant improvement in experiments with nonconvex classification and training of deep learning models. | We provide a new method that improves the state-of-the-art theoretical complexity of distributed optimization methods with compressed communication in the nonconvex regime. |
In reinforcement learning applications, agents usually need to deal with various input/output features when specified with different state and action spaces by their developers or physical restrictions, indicating re-training from scratch and considerable sample inefficiency, especially when agents follow similar solution steps to achieve tasks.
In this paper, we aim to transfer pre-trained skills to alleviate the above challenge. Specifically, we propose PILoT, i.e., Planning Immediate Landmarks of Targets. PILoT utilizes the universal decoupled policy optimization to learn a goal-conditioned state planner; then, we distill a goal-planner to plan immediate landmarks in a model-free style that can be shared among different agents. In our experiments, we show the power of PILoT on various transferring challenges, including few-shot transferring across action spaces and dynamics, from low-dimensional vector states to image inputs, from simple robot to complicated morphology; and we also illustrate PILoT provides a zero-shot transfer solution from a simple 2D navigation task to the harder Ant-Maze task. | We propose PILoT, a learning framework for transferring multi-task skills across agents. |
Federated domain adaptation (FDA) describes the setting where a set of source clients seek to optimize the performance of a target client. To be effective, FDA must address some of the distributional challenges of Federated learning (FL). For instance, FL systems exhibit distribution shifts across clients. Further, labeled data are not always available among the clients. To this end, we propose and compare novel approaches for FDA, combining the few labeled target samples with the source data when auxiliary labels are available to the clients. The in-distribution auxiliary information is included during local training to boost out-of-domain accuracy. Also, during fine-tuning, we devise a simple yet efficient gradient projection method to detect the valuable components from each source client model towards the target direction. The extensive experiments on medical imaging datasets show that our proposed framework significantly improves federated domain adaptation performance. | We leverage auxiliary information and propose gradient projection (GP) to tackle federated domain adaptation problem under weak supervision. |
Partial label learning (PLL) tackles the problem where each instance is associated with a set of candidate labels, only one of which is the ground-truth label. Most existing PLL approaches assume that both the training and test sets share an identical data distribution. However, this assumption does not hold in many real-world scenarios where the training and test data come from different distributions. In this paper, we formalize this learning scenario as a new problem called partial label unsupervised domain adaptation (PLUDA). To address this challenging PLUDA problem, we propose a novel Prototype Alignment based PLUDA method named PAPLUDA, which dynamically refines the pseudo-labels of instances from both the source and target domains by consulting the outputs of a teacher-student model in a moving-average manner, and bridges the cross-domain discrepancy through inter-domain class-prototype alignment. In addition, a teacher-student model based contrastive regularization is deployed to enhance prediction stability and hence improve the class-prototypes in both domains for PLUDA. Comprehensive experimental results demonstrate that PAPLUDA achieves state-of-the-art performance on the widely used benchmark datasets. | This is the first partial label learning method that handles partial label learning and unsupervised domain adaptation simultaneously. |
Meta-gradient reinforcement learning (RL) algorithms have substantially boosted the performance of RL agents by learning an adaptive return. All the existing algorithms adhere to the same reward learning principle, where the adaptive return is simply formulated in the form of expected cumulative rewards, upon which the policy and critic update rules are specified under well-adopted distance metrics. In this paper, we present a novel algorithm that builds on the success of meta-gradient RL algorithms and effectively improves such algorithms by following a simple recipe, i.e., going beyond the expected return to formulate and learn the return in a more expressive form, value distributions. To this end, we first formulate a distributional return that could effectively capture bootstrapping and discounting behaviors over distributions, to form an informative distributional return target in value update. Then we derive an efficient meta update rule to learn the adaptive distributional return with meta-gradients. For empirical evaluation, we first present an illustrative example on a toy two-color grid-world domain, which validates the benefit of learning distributional return over expectation; then we conduct extensive comparisons on a large-scale RL benchmark Atari 2600, where we confirm that our proposed method with distributional return works seamlessly well with the actor-critic framework and leads to state-of-the-art median human normalized score among meta-gradient RL literature. | A model-free meta gradient RL algorithm with distributional return |
Synthetic data generation (SDG) has become a popular approach to release private datasets.
In SDG, a generative model is fitted on the private real data, and samples drawn from the model are released as the protected synthetic data.
While real-world datasets usually consist of multiple tables with potential \emph{many-to-many} relationships (i.e.~\emph{many-to-many datasets}), recent research in SDG mostly focuses on modeling tables \emph{independently} or only considers generating datasets with special cases of many-to-many relationships such as \emph{one-to-many}.
In this paper, we first study challenges of building faithful generative models for many-to-many datasets, identifying limitations of existing methods.
We then present a novel factorization for many-to-many generative models, which leads to a scalable generation framework by combining recent results from random graph theory and representation learning.
Finally, we extend the framework to establish the notion of $(\epsilon,\delta)$-differential privacy.
Through a real-world dataset, we demonstrate that our method can generate synthetic datasets while preserving information within and across tables better than its closest competitor. | We synthesise datasets with many-to-many relationships by first generating the relationships via random graph generation and then generating the data attributes. |
Molecular representation pretraining is critical in various applications for drug and material discovery due to the limited number of labeled molecules, and most existing work focuses on pretraining on 2D molecular graphs. However, the power of pretraining on 3D geometric structures has been less explored. This is owing to the difficulty of finding a sufficient proxy task that can empower the pretraining to effectively extract essential features from the geometric structures. Motivated by the dynamic nature of 3D molecules, where the continuous motion of a molecule in the 3D Euclidean space forms a smooth potential energy surface, we propose GeoSSL, a 3D coordinate denoising pretraining framework to model such an energy landscape. Further by leveraging an SE(3)-invariant score matching method, we propose GeoSSL-DDM in which the coordinate denoising proxy task is effectively boiled down to denoising the pairwise atomic distances in a molecule. Our comprehensive experiments confirm the effectiveness and robustness of our proposed method. | We propose GeoSSL, a self-supervised learning method using the denoising distance matching for molecular goemetry pretraining. |
Recent works(Li et al., 2020, Wan et al., 2021) characterize an important mechanism of normalized model trained with SGD and WD (Weight Decay), called Spherical Motion Dynamics (SMD), confirming its widespread effects in practice. However, no theoretical study is available on the influence of SMD on the training process of normalized models in literature. In this work, we seek to understand the effect of SMD by theoretically analyzing a simple normalized model, named as Noisy Rayleigh Quotient (NRQ). On NRQ, We theoretically prove SMD can dominate the whole training process via controlling the evolution of angular update (AU), an essential feature of SMD. Specifically, we show: 1) within equilibrium state of SMD, the convergence rate and limiting risk of NRQ are mainly determined by the theoretical value of AU; and 2) beyond equilibrium state, the evolution of AU can interfere the optimization trajectory, causing odd phenomena such as ``escape'' behavior. We further show the insights drawn from NRQ is consistent with empirical observations in experiments on real datasets. We believe our theoretical results shed new light on the role of normalization techniques during the training of modern deep learning models. | Theoretical and empirical analysis on learning dynamics of neural network with normalization and weight decay. |
Originated as a philosophical quest, personality discerns how individuals differ from each other in terms of thinking, feeling, and behaving. Toward building social machines that work with humans on a daily basis, we are motivated to ask: (1) Do existing Large Language Models (LLMs) possess personalities, akin to their human counterparts? (2) If so, how can we evaluate them? (3) Further, given this evaluation framework, how can we induce a certain personality in a fully controllable fashion? To tackle these three questions, we propose the Machine Personality Inventory (MPI) dataset for evaluating the machine personality; MPI follows standardized personality tests, built upon the Big Five Personality Factors (Big Five) theory and personality assessment inventories. By evaluating models with MPI, we provide the first piece of evidence showing the existence of personality in LLMs. We further devise a Chain Prompting method to induce LLMs with a specific personality in a controllable manner, capable of producing diversified behaviors. We hope to shed light on future studies by adopting personality as the essential guide for various downstream tasks, building more human-like and in situ dialogue agents. | We propose the Machine Personality Inventory (MPI) dataset for evaluating the machine personality and devise a Chain Prompting method to induce the language model with a specific personality, capable of producing diversified behaviors. |
Neural Operators that directly learn mappings between function spaces have received considerable recent attention. Deep Operator Networks (DeepONets), a popular recent class of neural operators have shown promising preliminary results in approximating solution operators of parametric differential equations. Despite the universal approximation guarantees, there is yet no optimization convergence guarantee for DeepONets based on gradient descent (GD). In this paper, we establish such guarantees and show that over-parameterization based on wide layers provably helps. In particular, we present two types of optimization convergence analysis: first, for smooth activations, we bound the spectral norm of the Hessian of DeepONets and use the bound to show geometric convergence of GD based on restricted strong convexity (RSC); and second, for ReLU activations, we show the neural tangent kernel (NTK) of DeepONets at initialization is positive definite, which can be used with the standard NTK analysis to imply geometric convergence. Further, we present empirical results on three canonical operator learning problems: Antiderivative, Diffusion-Reaction equation, and Burger’s equation, and show that wider DeepONets lead to lower training loss on all the problems, thereby supporting the theoretical results | We show that stochastic gradient descent converges to the global minimum for a DeepONet model. |
This work presents a new dynamic and fully-connected layer (DFC) that generalizes existing layers and is free from hard inductive biases. Then, it describes how to factorize the DFC weights efficiently.
Using the Einstein convention as framework, we define the DFC as a fully connected layer with the weight tensor created as a function of the input. DFC is the non-linear extension of the most general case of linear layer for neural network, and therefore all major neural network layers, from convolution to self-attention, are particular cases of DFCs. A stack of DFCs interleaved by non-linearities defines a new super-class of neural networks: \emph{Formers}.
DFC has four major characteristics: it is Dynamic and Spatially Adaptive, it has a Global Receptive Field, and it mixes all the available channels' information.
In their complete form, DFCs are powerful layers free from hard inductive biases, but their use is limited in practice by their prohibitive computational cost. To overcome this limitation and deploy DFC in real computer-vision applications, we propose to use the CP decomposition, showing that it is possible to factorize the DFC layer into smaller, manageable blocks without losing any representational power. Finally, we propose ChoP'D Former, an architecture making use of a new decomposition of the DFC layer into five sequential operations, each incorporating one characteristic of the original DFC tensor. Chop'D Former leverages dynamic gating and integral image, achieves global spatial reasoning with constant time complexity, and has a receptive field that can adapt depending on the task. Extensive experiments demonstrate that our ChoP'D Former is competitive with state-of-the-art results on three well-known computer vision benchmarks, namely Large-Scale Classification, Object Detection, and Instance Segmentation, suppressing the need for expensive architecture search and hyperparameter optimization. | In this work, we unify prior methods and present a new efficient factorization for a general fully-connected and dynamic layer. |
Humans learn by interacting with their environments and perceiving the outcomes of their actions. A landmark in artificial intelligence has been the development of deep reinforcement learning (dRL) algorithms capable of doing the same in video games, on par with or better than humans. However, it remains unclear whether the successes of dRL models reflect advances in visual representation learning, the effectiveness of reinforcement learning algorithms at discovering better policies, or both. To address this question, we introduce the Learning Challenge Diagnosticator (LCD), a tool that separately measures the perceptual and reinforcement learning demands of a task. We use LCD to discover a novel taxonomy of challenges in the Procgen benchmark, and demonstrate that these predictions are both highly reliable and can instruct algorithmic development. More broadly, the LCD reveals multiple failure cases that can occur when optimizing dRL algorithms over entire video game benchmarks like Procgen, and provides a pathway towards more efficient progress. | Strategies for improving deep reinforcement learning agents can be predicted from their generalization performance. |
Tabular data synthesis is a long-standing research topic in machine learning. Many different methods have been proposed over the past decades, ranging from statistical methods to deep generative methods. However, it has not always been successful due to the complicated nature of real-world tabular data. In this paper, we present a new model named $\textbf{S}$core-based $\textbf{Ta}$bular data $\textbf{Sy}$nthesis ($\texttt{STaSy}$) and its training strategy based on the paradigm of score-based generative modeling. Despite the fact that score-based generative models have resolved many issues in generative models, there still exists room for improvement in tabular data synthesis. Our proposed training strategy includes a self-paced learning technique and a fine-tuning strategy, which further increases the sampling quality and diversity by stabilizing the denoising score matching training. Furthermore, we also conduct rigorous experimental studies in terms of the generative task trilemma: sampling quality, diversity, and time. In our experiments with 15 benchmark tabular datasets and 7 baselines, our method outperforms existing methods in terms of task-dependant evaluations and diversity.
| We design a score-based generative model for tabular data and apply two training strategies, including the self-paced learning and the proposed fine-tuning method, to stabilize the denoising score matching training. |
We propose an efficient design of Transformer-based models for multivariate time series forecasting and self-supervised representation learning. It is based on two key components: (i) segmentation of time series into subseries-level patches which are served as input tokens to Transformer; (ii) channel-independence where each channel contains a single univariate time series that shares the same embedding and Transformer weights across all the series. Patching design naturally has three-fold benefit: local semantic information is retained in the embedding; computation and memory usage of the attention maps are quadratically reduced given the same look-back window; and the model can attend longer history. Our channel-independent patch time series Transformer (PatchTST) can improve the long-term forecasting accuracy significantly when compared with that of SOTA Transformer-based models. We also apply our model to self-supervised pre-training tasks and attain excellent fine-tuning performance, which outperforms supervised training on large datasets. Transferring of masked pre-training performed on one dataset to other datasets also produces SOTA forecasting accuracy. | Channel-independent patch time series transformer works very well for long-term forecasting and representation learning. |
Since Batch Normalization was proposed, it has been commonly located in front of activation functions, as proposed by the original paper. Swapping the order, i.e., using Batch Normalization after activation functions, has also been attempted, but it is generally not much different from the conventional order when ReLU is used. However, in the case of bounded activation functions like Tanh, we discovered that the swapped order achieves considerably better performance on various benchmarks and architectures than the conventional order. We report this remarkable phenomenon and closely examine what contributes to this performance improvement in this paper. One noteworthy thing about swapped models is the extreme saturation of activation values, which is usually considered harmful. Looking at the output distribution of individual activation functions, we found that many of them are highly asymmetrically saturated. The experiments inducing a different degree of asymmetric saturation support the hypothesis that asymmetric saturation helps improve performance. In addition, we found that Batch Normalization after bounded activation functions has another important effect: it relocates the asymmetrically saturated output of activation functions near zero. This enables the swapped model to have higher sparsity, further improving performance. Extensive experiments with Tanh, LeLecun Tanh, and Softsign show that the swapped models achieve improved performance with a high degree of asymmetric saturation. | With bounded activation functions, using batch normalization after activation functions is better because of asymmetric saturation and sparsity. |
Transformer architectures have achieved great success in solving natural language tasks, which learn strong language representations from large-scale unlabeled texts. In this paper, we seek to go further beyond and explore a new logical inductive bias for better language representation learning. Logic reasoning is known as a formal methodology to reach answers from given knowledge and facts. Inspired by such a view, we develop a novel neural architecture named FOLNet (First-Order Logic Network), to encode this new inductive bias. We construct a set of neural logic operators as learnable Horn clauses, which are further forward-chained into a fully differentiable neural architecture (FOLNet). Interestingly, we find that the self-attention module in transformers can be composed by two of our neural logic operators, which probably explains their strong reasoning performance. Our proposed FOLNet has the same input and output interfaces as other pretrained models and thus could be pretrained/finetuned by using similar losses. It also allows FOLNet to be used in a plug-and-play manner when replacing other pretrained models. With our logical inductive bias, the same set of ``logic deduction skills'' learned through pretraining are expected to be equally capable of solving diverse downstream tasks. For this reason, FOLNet learns language representations that have much stronger transfer capabilities. Experimental results on several language understanding tasks show that our pretrained FOLNet model outperforms the existing strong transformer-based approaches. | We develop a novel neural architecture for learning language representations. |
Despite many advances in Graph Neural Networks (GNNs), their training strategies simply focus on minimizing a loss over nodes in a graph. However, such simplistic training strategies may be sub-optimal as they neglect that certain nodes are much harder to make accurate predictions on than others. Here we present TuneUp, a curriculum learning strategy for better training GNNs. Crucially, TuneUp trains a GNN in two stages. The first stage aims to produce a strong base GNN. Such base GNNs tend to perform well on head nodes (nodes with large degrees) but less so on tail nodes (nodes with small degrees). So, the second stage of TuneUp specifically focuses on improving prediction on tail nodes. Concretely, TuneUp synthesizes many additional supervised tail node data by dropping edges from head nodes and reusing the supervision on the original head nodes. TuneUp then minimizes the loss over the synthetic tail nodes to finetune the base GNN. TuneUp is a general training strategy that can be used with any GNN architecture and any loss, making TuneUp applicable to a wide range of prediction tasks. Extensive evaluation of TuneUp on two GNN architectures, three types of prediction tasks, and both inductive and transductive settings shows that TuneUp significantly improves the performance of the base GNN on tail nodes, while often even improving the performance on head nodes, which together leads up to 58.5% relative improvement in GNN predictive performance. Moreover, TuneUp significantly outperforms its variants without the two-stage curriculum learning, existing graph data augmentation techniques, as well as other specialized methods for tail nodes. | We develop a curriculum learning strategy to train GNNs with high generalization performance especially on tail nodes. |
Contrastive learning aims to extract distinctive features from data by finding an embedding representation where similar samples are close to each other, and different ones are far apart. We study how NNs generalize the concept of similarity in the presence of noise, investigating two phenomena: Double Descent (DD) behavior and online/offline correspondence. While DD examines how the network adjusts to the dataset during a long training time or by increasing the number of parameters, online/offline correspondence compares the network performances varying the quality (diversity) of the dataset. We focus on the simplest contrastive learning representative: Siamese Neural Networks (SNNs). We point out that SNNs can be affected by two distinct sources of noise: Pair Label Noise (PLN) and Single Label Noise (SLN). The effect of SLN is asymmetric, but it preserves similarity relations, while PLN is symmetric but breaks transitivity. We find that DD also appears in SNNs and is exacerbated by noise. We show that the dataset topology crucially affects generalization. While sparse datasets show the same performances under SLN and PLN for an equal amount of noise, SLN outperforms PLN in the overparametrized region in dense datasets. Indeed, in this regime, PLN similarity violation becomes macroscopical, corrupting the dataset to the point where complete overfitting cannot be achieved. We call this phenomenon Density-Induced Break of Similarity (DIBS). Probing the equivalence between online optimization and offline generalization in SNNs, we find that their correspondence breaks down in the presence of label noise for all the scenarios considered. | We investigate for the first time double descent and online/offline training in the context of similarity learning and find that the resulting learning model is heavily affected both by the topology of the dataset and noise. |
Siamese Networks are a popular self-supervised learning framework that learns useful representation without human supervision by encouraging representations to be invariant to distortions. Existing methods heavily rely on hand-crafted augmentations, which are not easily adapted to new domains. To explore a general-purpose or domain-agnostic siamese network, we investigate using masking as augmentations in siamese networks. Recently, masking for siamese networks has only been shown useful with transformer architectures, e.g. MSN and data2vec. In this work, we identify the underlying problems of masking for siamese networks with arbitrary backbones, including ConvNets. We propose an effective and general-purpose masking strategy and demonstrate its effectiveness on various siamese network frameworks. Our method generally improves siamese networks' performances in the few-shot image classification, and object detection tasks. | We propose a masking strategy for siamese networks with ConvNets. |
Nullspace of a linear mapping is the subspace which is mapped to the zero vector. For a linear map, adding an element of the nullspace to its input has no effect on the output of the mapping. We position this work as an exposition towards answering one simple question, ``Does a vision transformer have a non-trivial nullspace?" If TRUE, this would imply that adding elements from this non-trivial nullspace to an input will have no effect on the output of the network. This finding can eventually lead us closer to understanding the generalization properties of vision transformers. In this paper, we first demonstrate that provably a non-trivial nullspace exists for a particular class of vision transformers. This proof is drawn by simply computing the nullspace of the patch embedding matrices. We extend this idea to the non-linear layers of the vision transformer and show that it is possible to learn a non-linear counterpart to the nullspace via simple optimisations for any vision transformer. Subsequently, we perform studies to understand robustness properties of ViTs under nullspace noise. Under robustness, we investigate prediction stability, and (network and interpretation) fooling properties of the noise. Lastly, we provide image watermarking as an application of nullspace noise. | Our work highlights and discusses the concept of nullspace wrt vision transformers. |
Grokking, the unusual phenomenon for algorithmic datasets where generalization happens long after overfitting the training data, has remained elusive. We aim to understand grokking by analyzing the loss landscapes of neural networks, identifying the mismatch between training and test losses as the cause for grokking. We refer to this as the "LU mechanism" because training and test losses (against model weight norm) typically resemble "L" and "U", respectively. This simple mechanism can nicely explain many aspects of grokking: data size dependence, weight decay dependence, the emergence of representations, etc. Guided by the intuitive picture, we are able to induce grokking on tasks involving images, language and molecules, although the grokking signals are sometimes less dramatic. We attribute the dramatic nature of grokking for algorithmic datasets to representation learning. | We aim to understand grokking through the lens of neural loss landscapes, and show grokking can occur for various datasets beyond algorithmic datasets. |
Humans excel at lifelong learning, as the brain has evolved to be robust to distribution shifts and noise in our ever-changing environment. Deep neural networks (DNNs), however, exhibit catastrophic forgetting and the learned representations drift drastically as they encounter a new task. This alludes to a different error-based learning mechanism in the brain. Unlike DNNs, where learning scales linearly with the magnitude of the error, the sensitivity to errors in the brain decreases as a function of their magnitude. To this end, we propose "ESMER" which employs a principled mechanism to modulate error sensitivity in a dual-memory rehearsal-based system. Concretely, it maintains a memory of past errors and uses it to modify the learning dynamics so that the model learns more from small consistent errors compared to large sudden errors. We also propose "Error-Sensitive Reservoir Sampling" to maintain episodic memory, which leverages the error history to pre-select low-loss samples as candidates for the buffer, which are better suited for retaining information. Empirical results show that ESMER effectively reduces forgetting and abrupt drift in representations at the task boundary by gradually adapting to the new task while consolidating knowledge. Remarkably, it also enables the model to learn under high levels of label noise, which is ubiquitous in real-world data streams. | A novel method that employs a principled mechanism for modulating the error sensitivity in a dual-memory rehearsal-based system for effective continual learning |
The long-standing theory that a colour-naming system evolves under the dual pressure of efficient communication and perceptual mechanism is supported by more and more linguistic studies including the analysis of four decades’ diachronic data from the Nafaanra language. This inspires us to explore whether artificial intelligence could evolve and discover a similar colour-naming system via optimising the communication efficiency represented by high-level recognition performance. Here, we propose a novel colour quantisation transformer, CQFormer, that quantises colour space while maintaining the accuracy of machine recognition on the quantised image. Given an RGB image, the annotation branch maps it into an index map before generating the quantised image with a colour palette, meanwhile the palette branch utilises a key-point detection way to find proper colours in palette among whole colour space. By interacting with colour annotation, CQFormer is able to balance both the machine vision accuracy and colour perceptual structure such as distinct and stable colour distribution for discovered colour system. Very interestingly, we even observe the consistent evolution pattern between our artificial colour system and basic colour terms across human languages. Besides, our colour quantisation method also offers an efficient quantisation method that effectively compresses the image storage while maintaining a high performance in high-level recognition tasks such as classification and detection. Extensive experiments demonstrate the superior performance of our method with extremely low bit-rate colours. We will release the source code upon acceptance. | a new colour quantistaion transformer to artificially discover and evolve colour naming system similar in human language |
Recent Self-Supervised Learning (SSL) methods are able to learn feature representations that are invariant to different data augmentations, which can then be transferred to downstream tasks of interest. However, different downstream tasks require different invariances for their best performance, so the optimal choice of augmentations for SSL depends on the target task. In this paper, we aim to learn self-supervised features that generalize well across a variety of downstream tasks (e.g., object classification, detection and instance segmentation) without knowing any task information beforehand. We do so by Masked Augmentation Subspace Training (or MAST) to encode in the single feature space the priors from different data augmentations in a factorized way. Specifically, we disentangle the feature space into separate subspaces, each induced by a learnable mask that selects relevant feature dimensions to model invariance to a specific augmentation. We show the success of MAST in jointly capturing generalizable priors from different augmentations, using both unique and shared features across the subspaces. We further show that MAST benefits from uncertainty modeling to reweight ambiguous samples from strong augmentations that may cause similarity mismatch in each subspace. Experiments demonstrate that MAST consistently improves generalization on various downstream tasks, while being task-agnostic and efficient during SSL. We also provide interesting insights about how different augmentations are related and how uncertainty reflects learning difficulty. | Disentangled and uncertainty-aware learning of augmentation invariances during SSL improves generalization on downstream tasks |
Many policy gradient methods are variants of Actor-Critic (AC), where a value function (critic) is learned to facilitate updating the parameterized policy (actor). The update to the actor involves a log-likelihood update weighted by the action-values, with the addition of entropy regularization for soft variants. In this work, we explore an alternative update for the actor, based on an extension of the cross entropy method (CEM) to condition on inputs (states). The idea is to start with a broader policy and slowly concentrate around maximal actions, using a maximum likelihood update towards actions in the top percentile per state. The speed of this concentration is controlled by a proposal policy, that concentrates at a slower rate than the actor. We first provide a policy improvement result in an idealized setting, and then prove that our conditional CEM (CCEM) strategy tracks a CEM update per state, even with changing action-values. We empirically show that our Greedy AC algorithm, that uses CCEM for the actor update, performs better than Soft Actor-Critic and is much less sensitive to entropy-regularization. | We propose an alternative update for the actor in actor-critic algorithms that does not rely on entropy-regularization |
Diffusion models based on stochastic differential equations (SDEs) gradually perturb a data distribution $p(\mathbf{x})$ over time by adding noise to it. A neural network is trained to approximate the score $\nabla_\mathbf{x} \log p_t(\mathbf{x})$ at time $t$, which can be used to reverse the corruption process. In this paper, we focus on learning the score field that is associated with the time evolution according to a physics operator in the presence of natural non-deterministic physical processes like diffusion. A decisive difference to previous methods is that the SDE underlying our approach transforms the state of a physical system to another state at a later time. For that purpose, we replace the drift of the underlying SDE formulation with a differentiable simulator or a neural network approximation of the physics. At the core of our method, we optimize the so-called probability flow ODE to fit a training set of simulation trajectories inside an ODE solver and solve the reverse-time SDE for inference to sample plausible trajectories that evolve towards a given end state. We demonstrate the competitiveness of our approach for different challenging inverse problems. | We propose to learn score fields with a differentiable physics operator for natural non-deterministic physical processes like diffusion in order to solve inverse problems and obtain their posterior distribution. |
In distributed computing, slower nodes (stragglers) usually become a bottleneck. Gradient Coding (GC), introduced by Tandon et al., is an efficient technique that uses principles of error-correcting codes to distribute gradient computation in the presence of stragglers. In this paper, we consider the distributed computation of a sequence of gradients $\{g(1),g(2),\ldots,g(J)\}$, where processing of each gradient $g(t)$ starts in round-$t$ and finishes by round-$(t+T)$. Here $T\geq 0$ denotes a delay parameter. For the GC scheme, coding is only across computing nodes and this results in a solution where $T=0$. On the other hand, having $T>0$ allows for designing schemes which exploit the temporal dimension as well. In this work, we propose two schemes that demonstrate improved performance compared to GC. Our first scheme combines GC with selective repetition of previously unfinished tasks and achieves improved straggler mitigation. In our second scheme, which constitutes our main contribution, we apply GC to a subset of the tasks and repetition for the remainder of the tasks. We then multiplex these two classes of tasks across workers and rounds in an adaptive manner, based on past straggler patterns. Using theoretical analysis, we demonstrate that our second scheme achieves significant reduction in the computational load. In our experiments, we study a practical setting of concurrently training multiple neural networks over an AWS Lambda cluster involving 256 worker nodes, where our framework naturally applies. We demonstrate that the latter scheme can yield a 16\% improvement in runtime over the baseline GC scheme, in the presence of naturally occurring, non-simulated stragglers.
| We propose to improve gradient coding by exploiting the temporal dimension while training deep learning models in distributed cloud systems. |
Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Several techniques have achieved learning with no CF. However, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel and simple technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction of capacity usage. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while preventing CF. | This work aims to (1) overcome catastrophic forgetting, (2) encourage knowledge transfer, and (3) tackle the capacity problem in continual learning. |
Decentralized learning algorithms enable the training of deep learning models over large distributed datasets generated at different devices and locations, without the need for a central server. In practical scenarios, the distributed datasets can have significantly different data distributions across the agents. The current state-of-the-art decentralized algorithms mostly assume the data distributions to be Independent and Identically Distributed (IID). This paper focuses on improving decentralized learning over non-IID data distributions with minimal compute and memory overheads. We propose Neighborhood Gradient Clustering (NGC), a novel decentralized learning algorithm that modifies the local gradients of each agent using self- and cross-gradient information. Cross-gradients for a pair of neighboring agents are the derivatives of the model parameters of an agent with respect to the dataset of the other agent. In particular, the proposed method replaces the local gradients of the model with the weighted mean of the self-gradients, model-variant cross-gradients (derivatives of the received neighbors’ model parameters with respect to the local dataset - computed locally), and data-variant cross-gradients (derivatives of the local model with respect to its neighbors’ datasets - received through communication). The data-variant cross-gradients are aggregated through an additional communication round without breaking the privacy constraints of the decentralized setting. Further, we present CompNGC, a compressed version of NGC that reduces the communication overhead by $32 \times$ by compressing the cross-gradients. We demonstrate the empirical convergence and efficiency of the proposed technique over non-IID data distributions sampled from the CIFAR-10 dataset on various model architectures and graph topologies. Our experiments demonstrate that NGC and CompNGC outperform the existing state-of-the-art (SoTA) decentralized learning algorithm over non-IID data by $1-5\%$ with significantly less compute and memory requirements. Further, we also show that the proposed NGC method outperforms the baseline by $5-40\%$ with no additional communication. | Proposed a novel decentralized learning algorithm to improve the performance over non-IID data distributions through manipulation of local-gradients |
In reward-free reinforcement learning (RL), an agent explores the environment first without any reward information, in order to achieve certain learning goals afterwards for any given reward. In this paper we focus on reward-free RL under low-rank MDP models, in which both the representation and linear weight vectors are unknown. Although various algorithms have been proposed for reward-free low-rank MDPs, the corresponding sample complexity is still far from being satisfactory. In this work, we first provide the first known sample complexity lower bound that holds for any algorithm under low-rank MDPs. This lower bound implies it is strictly harder to find a near-optimal policy under low-rank MDPs than under linear MDPs. We then propose a novel model-based algorithm, coined RAFFLE, and show it can both find an $\epsilon$-optimal policy and achieve an $\epsilon$-accurate system identification via reward-free exploration, with a sample complexity significantly improving the previous results. Such a sample complexity matches our lower bound in the dependence on $\epsilon$, as well as on $K$ {in the large $d$ regime}, where $d$ and $K$ respectively denote the representation dimension and action space cardinality. Finally, we provide a planning algorithm (without further interaction with true environment) for RAFFLE to learn a near-accurate representation, which is the first known representation learning guarantee under the same setting.
| We propose a novel reward free reinforcement learning algorithm under low-rank MDPs, which improves the sample complexity of previous work. We also provide a lower bound. Finally we study representation learning via reward free reinforement learning. |
We proposed Class-Informed Variational Autoencoder (CI-VAE) to enable interpolation between arbitrary pairs of observations of the same class. CI-VAE combines the general VAE architecture with a linear discriminator layer on the latent space to enforce the construction of a latent space such that observations from different classes are linearly separable. In conventional VAEs, class overlapping on the latent space usually occurs. However, in CI-VAE, the enforced linear separability of classes on the latent space allows for robust latent-space linear traversal and data generation between two arbitrary observations of the same class. Class-specific data interpolation has extensive potential applications in science, particularly in biology, such as uncovering the biological trajectory of diseases or cancer. We used the MNIST dataset of handwritten digits as a case study to compare the performance of CI-VAE and VAE in class-specific data augmentation. We showed that CI-VAE significantly improved class-specific linear traversal and data augmentation compared with VAE while maintaining comparable reconstruction error. In a study of Colon cancer genomics data, we showed that the interpolation between normal cells and tumor cells using CI-VAE may enhance our understanding of cancer development. | A deep learning framework for interpolations in high-dimensional data |
Gradient flows are differential equations that minimize an energy functional and constitute the main descriptors of physical systems. We apply this formalism to Graph Neural Networks (GNNs) to develop new frameworks for learning on graphs as well as provide a better theoretical understanding of existing ones. We derive GNNs as a gradient flow equation of a parametric energy that provides a physics-inspired interpretation of GNNs as learning particle dynamics in the feature space. In particular, we show that in graph convolutional models (GCN), the positive/negative eigenvalues of the channel mixing matrix correspond to attractive/repulsive forces between adjacent features. We rigorously prove how the channel-mixing can learn to steer the dynamics towards low or high frequencies, which allows to deal with heterophilic graphs. We show that the same class of energies is decreasing along a larger family of GNNs; albeit not gradient flows, they retain their inductive bias. We experimentally evaluate an instance of the gradient flow framework that is principled, more efficient than GCN, and achieves competitive performance on graph datasets of varying homophily often outperforming recent baselines specifically designed to target heterophily. | We apply the gradient flow formalism to GNNs to both develop new frameworks and provide a better theoretical understanding of existing ones. |
Transformers have become a preferred tool for modeling sequential data. Many studies of using Transformers for long sequence modeling focus on reducing computational complexity. They usually exploit the low-rank structure of data and approximate a long sequence by a sub-sequence. One challenge with such approaches is how to make an appropriate tradeoff between information preserving and noise reduction: the longer the sub-sequence used to approximate the long sequence, the better the information is preserved but at a price of introducing more noise into the model and of course more computational costs. We propose skeleton transformer, SKTformer for short, an efficient transformer architecture that effectively addresses the tradeoff. It introduces two mechanisms to effectively reduce the impact of noise while still keeping the computation linear to the sequence length: a smoothing block to mix information over long sequences and a matrix sketch method that simultaneously selects columns and rows from the input matrix. We verify the effectiveness of SKTformer both theoretically and empirically. Extensive studies over both Long Range Arena (LRA) datasets and six time-series forecasting show that SKTformer significantly outperforms both villain Transformer and other state-of-the-art variants of Transformer. Code is available at
https://anonymous.4open.science/r/SKTFormer-B33B/ | We design an efficient Transformer model for long sequence data |
Recently, state space models (SSMs) have shown promising results on sequence modeling tasks. However, a potential challenge of existing works is that SSMs are usually introduced or initialized in a homogeneous way, encouraging the model to only capture similar temporal dynamics on different features. In this paper, we propose a multi-head state space model (MSSM), in which parallel heads are introduced to learn different temporal dynamics on sequence data. Furthermore, we propose a novel variant of the Transformer, referred to as the Stateformer, which combines MSSMs with attention. Experiments on large-scale automatic speech recognition (ASR) and language modeling tasks show the MSSM outperforming a range of attention-based baselines. The Stateformer further improves performance, achieving the state-of-the-art performance on the LibriSpeech ASR task. | We develop a novel multi-head state space model as a replacement and/or complement to attention, achieving state-of-the-art performance in speech recognition and masked language modeling. |
Distributional reinforcement learning~(RL) is a class of state-of-the-art algorithms that estimate the entire distribution of the total return rather than its expected value alone. The theoretical advantages of distributional RL over expectation-based RL remain elusive, despite the remarkable performance of distributional RL. Our work attributes the superiority of distributional RL to its regularization effect stemming from the value distribution information regardless of only its expectation. We decompose the value distribution into its expectation and the remaining distribution part using a variant of the gross error model in robust statistics. Hence, distributional RL has an additional benefit over expectation-based RL thanks to the impact of a \textit{risk-sensitive entropy regularization} within the Neural Fitted Z-Iteration framework. Meanwhile, we investigate the role of the resulting regularization in actor-critic algorithms by bridging the risk-sensitive entropy regularization of distributional RL and the vanilla entropy in maximum entropy RL. It reveals that distributional RL induces an augmented reward function, which promotes a risk-sensitive exploration against the intrinsic uncertainty of the environment. Finally, extensive experiments verify the importance of the regularization effect in distributional RL, as well as the mutual impacts of different entropy regularizations. Our study paves the way towards a better understanding of distributional RL, especially when looked at through a regularization lens. | We interpret distributional reinforcement learning from the perspectives of regularization. |
Normalizing flow is a class of deep generative models for efficient sampling and density estimation. In practice, the flow often appears as a chain of invertible neural network blocks. To facilitate training, past works have regularized flow trajectories and designed special network architectures. The current paper develops a neural ODE flow network inspired by the Jordan-Kinderleherer-Otto (JKO) scheme, which allows an efficient \textit{block-wise} training procedure: as the JKO scheme unfolds the dynamic of gradient flow, the proposed model naturally stacks residual network blocks one-by-one and reduces the memory load as well as the difficulty of training deep networks. We also develop an adaptive time-reparametrization of the flow network with a progressive refinement of the trajectory in probability space, which improves the optimization efficiency and model accuracy in practice.
On high-dimensional generative tasks for tabular data, JKO-Flow can process larger data batches and perform competitively as or better than continuous and discrete flow models, using 10X less number of iterations (e.g., batches) and significantly less time per iteration. | We propose JKO-Flow to train normalizing flow neural ODE model block-wise with time reparametrization, and experimentally show JKO-Flow reaches competitive performance while greatly reduce computation |
Muscle-actuated organisms are capable of learning an unparalleled diversity of dexterous movements despite their vast amount of muscles.
Reinforcement learning (RL) on large musculoskeletal models, however, has not been able to show similar performance.
We conjecture that ineffective exploration in large overactuated action spaces is a key problem.
This is supported by the finding that common exploration noise strategies are inadequate in synthetic examples of overactuated systems.
We identify differential extrinsic plasticity (DEP), a method from the domain of self-organization, as being able to induce state-space covering exploration within seconds of interaction.
By integrating DEP into RL, we achieve fast learning of reaching and locomotion in musculoskeletal systems, outperforming current approaches in all considered tasks in sample efficiency and robustness. | A technique from the self-organization literature is used to improve performance of RL agents on overactuated systems with up to 120 muscle actuators. |
We present a novel neural network architecture using self-attention, the Wavefunction Transformer (PsiFormer), which can be used as an approximation (or "Ansatz") for solving the many-electron Schrödinger equation, the fundamental equation for quantum chemistry and material science. This equation can be solved *from first principles*, requiring no external training data. In recent years, deep neural networks like the FermiNet and PauliNet have been used to significantly improve the accuracy of these first-principle calculations, but they lack an attention-like mechanism for gating interactions between electrons. Here we show that the PsiFormer can be used as a drop-in replacement for these other neural networks, often dramatically improving the accuracy of the calculations. On larger molecules especially, the ground state energy can be improved by dozens of kcal/mol, a qualitative leap over previous methods. This demonstrates that self-attention networks can learn complex quantum mechanical correlations between electrons, and are a promising route to reaching unprecedented accuracy in chemical calculations on larger systems. | We use a novel self-attention neural network to make quantum chemistry calculations from first principles much more accurate. |
Click-through rate (CTR) prediction plays important role in the advertisement, recommendation, and retrieval applications. Given the feature set, how to fully utilize the information from the feature set is an active topic in deep CTR model designs. There are several existing deep CTR works focusing on feature interactions, feature attentions, and so on. They attempt to capture high-order feature interactions to enhance the generalization ability of deep CTR models. However, these works either suffer from poor high-order feature interaction modeling using DNN or ignore the balance between generalization and memorization during the recommendation. To mitigate these problems, we propose an adaptive feature fusion framework called MaskFusion, to additionally capture the explicit interactions between the input feature and the existing deep part structure of deep CTR models dynamically, besides the common feature interactions proposed in existing works. MaskFusion is an instance-aware feature augmentation method, which makes deep CTR models more personalized by assigning each feature with an instance-adaptive mask and fusing each feature with each hidden state vector in the deep part structure. MaskFusion can also be integrated into any existing deep CTR models flexibly. MaskFusion achieves state-of-the-art (SOTA) performance on all seven benchmarks deep CTR models with three public datasets. | Feature Augmentation via Adaptive Mask Fusion |
In this paper, we consider the task of unsupervised object discovery in videos. Previous works have shown promising results via processing optical flows to segment objects. However, taking flow as input brings about two drawbacks. First, flow cannot capture sufficient cues when objects remain static or partially occluded. Second, it is challenging to establish temporal coherency from flow-only input, due to the missing texture information. To tackle these limitations, we propose a model for directly processing consecutive RGB frames, and infer the optical flow between any pair of frames using a layered representation, with the opacity channels being treated as the segmentation. Additionally, to enforce object permanence, we apply temporal consistency loss on the inferred masks from randomly-paired frames, which refer to the motions at different paces, and encourage the model to segment the objects even if they may not move at the current time point. Experimentally, we demonstrate superior performance over previous state-of-the-art methods on three public video segmentation datasets (DAVIS2016, SegTrackv2, and FBMS-59), while being computationally efficient by avoiding the overhead of computing optical flow as input. | We propose a motion-inductive model through directly processing consecutive RGB frames to segment the foreground objects and train it by flow reconstruction between pairwise frames, i.e. without any mask annotations. |
Despite the recent advances in communication-efficient distributed bandit learning, most existing solutions are restricted to parametric models, e.g., linear bandits and generalized linear bandits (GLB). In comparison, kernel bandits, which search for non-parametric functions in a reproducing kernel Hilbert space (RKHS), offer higher modeling capacity. But the only existing work in distributed kernel bandits adopts a synchronous communication protocol, which greatly limits its practical use (e.g., every synchronization step requires all clients to participate and wait for data exchange).
In this paper, in order to improve the robustness against delays and unavailability of clients that are common in practice, we propose the first asynchronous solution based on approximated kernel regression for distributed kernel bandit learning. A set of effective treatments are developed to ensure approximation quality and communication efficiency. Rigorous theoretical analysis about the regret and communication cost is provided; and extensive empirical evaluations demonstrate the effectiveness of our solution. | We propose and analyze a communication efficient asynchronous Kernel UCB algorithm with Nystrom approximation. |
This work pursues the optimization of over-parameterized deep models for superior training efficiency and test performance. We first theoretically emphasize the importance of two properties of over-parameterized models, i.e., the convergence gap and the generalization gap. Subsequent analyses unveil that these two gaps can be upper-bounded by the ratio of the Lipschitz constant and the Polyak-{\L}ojasiewicz (PL) constant, a crucial term abbreviated as the \emph{condition number}. Such discoveries have led to a structured pruning method with a novel pruning criterion. That is, we devise a gating network that dynamically detects and masks out those poorly-behaved nodes of a deep model during the training session. To this end, this gating network is learned via minimizing the \emph{condition number} of the target model, and this process can be implemented as an extra regularization loss term. Experimental studies demonstrate that the proposed method outperforms the baselines in terms of both training efficiency and test performance, exhibiting the potential of generalizing to a variety of deep network architectures and tasks. | This work proposes a new regularized risk minimization for over-parameterized models with a novel PL regularization and implements it via network pruning guided by PL-based condition number. |
We study reward-free reinforcement learning with linear function approximation for episodic Markov decision processes (MDPs). In this setting, an agent first interacts with the environment without accessing the reward function in the exploration phase. In the subsequent planning phase, it is given a reward function and asked to output an $\epsilon$-optimal policy. We propose a novel algorithm LSVI-RFE under the linear MDP setting, where the transition probability and reward functions are linear in a feature mapping. We prove an $\widetilde{O}(H^{4} d^{2}/\epsilon^2)$ sample complexity upper bound for LSVI-RFE, where $H$ is the episode length and $d$ is the feature dimension. We also establish a sample complexity lower bound of $\Omega(H^{3} d^{2}/\epsilon^2)$. To the best of our knowledge, LSVI-RFE is the first computationally efficient algorithm that achieves the minimax optimal sample complexity in linear MDP settings up to an $H$ and logarithmic factors. Our LSVI-RFE algorithm is based on a novel variance-aware exploration mechanism to avoid overly-conservative exploration in prior works. Our sharp bound relies on the decoupling of UCB bonuses during two phases, and a Bernstein-type self-normalized bound, which remove the extra dependency of sample complexity on $H$ and $d$, respectively. | We propose a computationally-efficient algorithm for reward-free exploration in linear MDPs reaching a minimax optimal sample complexity up to an $H$ and logarithm factor. |
Noise Contrastive Estimation (NCE) is a popular approach for learning probability density functions parameterized up to a constant of proportionality. The main idea is to design a classification problem for distinguishing training data from samples from an (easy-to-sample) noise distribution $q$, in a manner that avoids having to calculate a partition function. It is well-known that the choice of $q$ can severely impact the computational and statistical efficiency of NCE. In practice, a common choice for $q$ is a Gaussian which matches the mean and covariance of the data.
In this paper, we show that such a choice can result in an exponentially bad (in the ambient dimension) conditioning of the Hessian of the loss - even for very simple data distributions. As a consequence, both the statistical and algorithmic complexity for such a choice of $q$ will be problematic in practice - suggesting that more complex noise distributions are essential to the success of NCE. | We show that using Gaussians as the noise distribution in Noise Contrastive Estimation can lead to exponentially bad statistical and algorithmic complexity. |
While deep learning-based methods for blind face restoration have achieved unprecedented success, they still suffer from two major limitations. First, most of them deteriorate when facing complex degradations out of their training data. Second, these methods require multiple constraints, e.g., fidelity, perceptual, and adversarial losses, which requires laborious hyper-parameters tuning to stabilize and balance their influences. In this work, we propose a novel method named DifFace, being able to cope with unseen and complex degradations more gracefully without complicated loss designs. The key of our method is to establish a posterior distribution from the observed low-quality (LQ) image to its high-quality (HQ) counterpart. In particular, we design a transition distribution from the LQ image to the intermediate state of a pre-trained diffusion model and then gradually transmit from this intermediate state to the HQ target by recursively applying a pre-trained diffusion model. The transition distribution only relies on a restoration backbone that is trained with L2 loss on some synthetic data, which favorably avoids the cumbersome training process in existing methods. Moreover, the transition distribution is capable of contracting the error of the restoration backbone and thus makes our method more robust to unknown degradations. Comprehensive experiments show that DifFace is superior to current state-of-the-art methods, especially in cases with severe degradations. Code and model will be released. | We propose a new blind face restoration method that consists of an error compressor and a Markov chain partially borrowed from a pre-trained diffusion model. |
Meta-learning or learning to learn is a popular approach for learning new tasks with limited data (i.e., few-shot learning) by leveraging the commonalities among different tasks. However, meta-learned models can perform poorly when context data is limited, or when data is drawn from an out-of-distribution (OoD) task. Especially in safety-critical settings, this necessitates an uncertainty-aware approach to meta-learning. In addition, the often multimodal nature of task distributions can pose unique challenges to meta-learning methods. In this work, we present UnLiMTD (Uncertainty-aware meta-Learning for Multimodal Task Distributions), a novel method for meta-learning that (1) makes probabilistic predictions on in-distribution tasks efficiently, (2) is capable of detecting OoD context data at test time, and (3) performs on heterogeneous, multimodal task distributions. To achieve this goal, we take a probabilistic perspective and train a parametric, tuneable distribution over tasks on the meta-dataset. We construct this distribution by performing Bayesian inference on a linearized neural network, leveraging Gaussian process theory. We demonstrate that UnLiMTD’s predictions compare to, and outperform in most cases, the standard baselines, especially in the low-data regime. Furthermore, we show that UnLiMTD is effective in detecting data from OoD tasks. Finally, we confirm that both of these findings continue to hold in the multimodal task-distribution setting. | We present a novel meta-learning algorithm that makes probabilistic predictions efficiently, detects out-of-distribution context data, and performs well on heterogeneous, multimodal task distributions. |
Graph contrastive learning (GCL), as an emerging self-supervised learning technique on graphs, aims to learn representations via instance discrimination. Its performance heavily relies on graph augmentation to reflect invariant patterns that are robust to small perturbations; yet it still remains unclear about what graph invariance GCL should capture. Recent studies mainly perform topology augmentations in a uniformly random manner in the spatial domain, ignoring its influence on the intrinsic structural properties embedded in the spectral domain. In this work, we aim to find a principled way for topology augmentations by exploring the invariance of graphs from the spectral perspective. We develop spectral augmentation which guides topology augmentations by maximizing the spectral change. Extensive experiments on both graph and node classification tasks demonstrate the effectiveness of our method in self-supervised representation learning. The proposed method also brings promising generalization capability in transfer learning, and is equipped with intriguing robustness property under adversarial attacks. Our study sheds light on a general principle for graph topology augmentation. | We propose a novel spectral augmentation method which uses graph spectrum to capture structural properties and guide topology augmentations for graph self-supervised learning. |
We study how to transfer representations pretrained on source tasks to target tasks in visual percept based RL. We analyze two popular approaches: freezing or finetuning the pretrained representations. Empirical studies on a set of popular tasks reveal several properties of pretrained representations. First, finetuning is required even when pretrained representations perfectly capture the information required to solve the target task. Second, finetuned representations improve learnability and are more robust to noise.
Third, pretrained bottom layers are task-agnostic and readily transferable to new tasks, while top layers encode task-specific information and require adaptation. Building on these insights, we propose a self-supervised objective that \emph{clusters representations according to the policy they induce}, as opposed to traditional representation similarity measures which are policy-agnostic (\eg Euclidean norm, cosine similarity). Together with freezing the bottom layers, this objective results in significantly better representation than frozen, finetuned, and self-supervised alternatives on a wide range of benchmarks. | We study the transfer of visual representations in RL, show that they can be partially frozen, and propose a self-supervised method to accelerate their finetuning. |
Quantifying similarity between neural representations---e.g. hidden layer activation vectors---is a perennial problem in deep learning and neuroscience research. Existing methods compare deterministic responses (e.g. artificial networks that lack stochastic layers) or averaged responses (e.g., trial-averaged firing rates in biological data). However, these measures of _deterministic_ representational similarity ignore the scale and geometric structure of noise, both of which play important roles in neural computation. To rectify this, we generalize previously proposed shape metrics (Williams et al. 2021) to quantify differences in _stochastic_ representations. These new distances satisfy the triangle inequality, and thus can be used as a rigorous basis for many supervised and unsupervised analyses. Leveraging this novel framework, we find that the stochastic geometries of neurobiological representations of oriented visual gratings and naturalistic scenes respectively resemble untrained and trained deep network representations. Further, we are able to more accurately predict certain network attributes (e.g. training hyperparameters) from its position in stochastic (versus deterministic) shape space. | Representational dissimilarity metrics that account for noise geometry in biological and artificial neural responses. |