text
stringlengths 23
5.09k
| label
class label 3
classes |
|---|---|
Title: MUSE: Multi-Scale Attention Model for Sequence to Sequence Learning. Abstract: Transformers have achieved state-of-the-art results on a variety of natural language processing tasks.
Despite good performance, Transformers are still weak in long sentence modeling where the global attention map is too dispersed to capture valuable information.
In such case, the local/token features that are also significant to sequence modeling are omitted to some extent.
To address this problem, we propose a Multi-scale attention model (MUSE) by concatenating attention networks with convolutional networks and position-wise feed-forward networks to explicitly capture local and token features. Considering the parameter size and computation efficiency, we re-use the feed-forward layer in the original Transformer and adopt a lightweight dynamic convolution as implementation.
Experimental results show that the proposed model achieves substantial performance improvements over Transformer, especially on long sentences, and pushes the state-of-the-art from 35.6 to 36.2 on IWSLT 2014 German to English translation task, from 30.6 to 31.3 on IWSLT 2015 English to Vietnamese translation task. We also reach the state-of-art performance on WMT 2014 English to French translation dataset, with a BLEU score of 43.2. | 2withdrawn
|
Title: Zero-Shot Recognition through Image-Guided Semantic Classification. Abstract: We present a new visual-semantic embedding method for generalized zero-shot learning. Existing embedding-based methods aim to learn the correspondence between an image classifier (visual representation) and its class prototype (semantic representation) for each class. Inspired by the binary relevance method for multi-label classification, we learn the mapping between an image and its semantic classifier. Given an input image, the proposed Image-Guided Semantic Classification (IGSC) method creates a label classifier, being applied to all label embeddings to determine whether a label belongs to the input image. Therefore, a semantic classifier is image conditioned and is generated during inference. We also show that IGSC is a unifying framework for two state-of-the-art deep-embedding methods. We validate our approach with four standard benchmark datasets.
| 0reject
|
Title: Not-So-Random Features. Abstract: We propose a principled method for kernel learning, which relies on a Fourier-analytic characterization of translation-invariant or rotation-invariant kernels. Our method produces a sequence of feature maps, iteratively refining the SVM margin. We provide rigorous guarantees for optimality and generalization, interpreting our algorithm as online equilibrium-finding dynamics in a certain two-player min-max game. Evaluations on synthetic and real-world datasets demonstrate scalability and consistent improvements over related random features-based methods. | 1accept
|
Title: DictFormer: Tiny Transformer with Shared Dictionary. Abstract: We introduce DictFormer with the efficient shared dictionary to provide a compact, fast, and accurate transformer model. DictFormer significantly reduces the redundancy in the transformer's parameters by replacing the prior transformer's parameters with a compact, shared dictionary, few unshared coefficients, and indices. Also, DictFormer enables faster computations since expensive weights multiplications are converted into cheap shared look-ups on dictionary and few linear projections. Training dictionary and coefficients are not trivial since indices used for looking up dictionary are not differentiable. We adopt a sparse-constraint training with $l_1\,\,norm$ relaxation to learn coefficients and indices in DictFormer. DictFormer is flexible to support different model sizes by dynamically changing dictionary size. Compared to existing lightweight Transformers, DictFormer consistently reduces model size over Transformer on multiple tasks, e.g., machine translation, abstractive summarization, and language modeling. Extensive experiments show that DictFormer reduces $3.6\times$ to $8.9\times$ model size with similar accuracy over multiple tasks, compared to Transformer. | 1accept
|
Title: The Impact of the Mini-batch Size on the Dynamics of SGD: Variance and Beyond. Abstract: We study mini-batch stochastic gradient descent (SGD) dynamics under linear regression and deep linear networks by focusing on the variance of the gradients only given the initial weights and mini-batch size, which is the first study of this nature. In the linear regression case, we show that in each iteration the norm of the gradient is a decreasing function of the mini-batch size $b$ and thus the variance of the stochastic gradient estimator is a decreasing function of $b$. For deep neural networks with $L_2$ loss we show that the variance of the gradient is a polynomial in $1/b$. The results theoretically back the important intuition that smaller batch sizes yield larger variance of the stochastic gradients and lower loss function values which is a common believe among the researchers. The proof techniques exhibit a relationship between stochastic gradient estimators and initial weights, which is useful for further research on the dynamics of SGD. We empirically provide insights to our results on various datasets and commonly used deep network structures. We further discuss possible extensions of the approaches we build in studying the generalization ability of the deep learning models. | 0reject
|
Title: Beyond Winning and Losing: Modeling Human Motivations and Behaviors with Vector-valued Inverse Reinforcement Learning. Abstract: In recent years, reinforcement learning methods have been applied to model gameplay with great success, achieving super-human performance in various environments, such as Atari, Go and Poker.
However, those studies mostly focus on winning the game and have largely ignored the rich and complex human motivations, which are essential for understanding the agents' diverse behavior.
In this paper, we present a multi-motivation behavior modeling which investigates the multifaceted human motivations and models the underlying value structure of the agents.
Our approach extends inverse RL to the vectored-valued setting which imposes a much weaker assumption than previous studies.
The vectorized rewards incorporate Pareto optimality, which is a powerful tool to explain a wide range of behavior by its optimality.
For practical assessment, our algorithm is tested on the World of Warcraft Avatar History dataset spanning three years of the gameplay.
Our experiments demonstrate the improvement over the scalarization-based methods on real-world problem settings. | 0reject
|
Title: Deep Online Learning Via Meta-Learning: Continual Adaptation for Model-Based RL. Abstract: Humans and animals can learn complex predictive models that allow them to accurately and reliably reason about real-world phenomena, and they can adapt such models extremely quickly in the face of unexpected changes. Deep neural network models allow us to represent very complex functions, but lack this capacity for rapid online adaptation. The goal in this paper is to develop a method for continual online learning from an incoming stream of data, using deep neural network models. We formulate an online learning procedure that uses stochastic gradient descent to update model parameters, and an expectation maximization algorithm with a Chinese restaurant process prior to develop and maintain a mixture of models to handle non-stationary task distributions. This allows for all models to be adapted as necessary, with new models instantiated for task changes and old models recalled when previously seen tasks are encountered again. Furthermore, we observe that meta-learning can be used to meta-train a model such that this direct online adaptation with SGD is effective, which is otherwise not the case for large function approximators. We apply our method to model-based reinforcement learning, where adapting the predictive model is critical for control; we demonstrate that our online learning via meta-learning algorithm outperforms alternative prior methods, and enables effective continuous adaptation in non-stationary task distributions such as varying terrains, motor failures, and unexpected disturbances. | 1accept
|
Title: Task Conditioned Stochastic Subsampling. Abstract: Deep Learning algorithms are designed to operate on huge volumes of high dimensional data such as images. In order to reduce the volume of data these algorithms must process, we propose a set-based two-stage end-to-end neural subsampling model that is jointly optimized with an \textit{arbitrary} downstream task network such as a classifier. In the first stage, we efficiently subsample \textit{candidate elements} using conditionally independent Bernoulli random variables, followed by conditionally dependent autoregressive subsampling of the candidate elements using Categorical random variables in the second stage. We apply our method to feature and instance selection and show that our method outperforms the relevant baselines under very low subsampling rates on many tasks including image classification, image reconstruction, function reconstruction and few-shot classification. Additionally, for nonparametric models such as Neural Processes that require to leverage whole training data at inference time, we show that our method enhances the scalability of these models. To ensure easy reproducibility, we provide source code in the \textbf{Supplementary Material}.
| 0reject
|
Title: Automatic Portrait Video Matting via Context Motion Network. Abstract: Automatic portrait video matting is an under-constrained problem. Most state-of-the-art methods only exploit the semantic information and process each frame individually. Their performance is compromised due to the lack of temporal information between the frames. To solve this problem, we explore the optical flow between video frames for the automatic portrait video matting. Specifically, we propose the context motion network to leverage semantic information and motion information. To capture the motion information, we estimate the optical flow and design a context-motion updating operator to integrate features between frames recurrently. Our experiments show that our network outperforms state-of-the-art matting methods significantly on the Video240K SD dataset. | 2withdrawn
|
Title: Feature prioritization and regularization improve standard accuracy and adversarial robustness. Abstract: Adversarial training has been successfully applied to build robust models at a certain cost. While the robustness of a model increases, the standard classification accuracy declines. This phenomenon is suggested to be an inherent trade-off. We propose a model that employs feature prioritization by a nonlinear attention module and $L_2$ feature regularization to improve the adversarial robustness and the standard accuracy relative to adversarial training. The attention module encourages the model to rely heavily on robust features by assigning larger weights to them while suppressing non-robust features. The regularizer encourages the model to extracts similar features for the natural and adversarial images, effectively ignoring the added perturbation. In addition to evaluating the robustness of our model, we provide justification for the attention module and propose a novel experimental strategy that quantitatively demonstrates that our model is almost ideally aligned with salient data characteristics. Additional experimental results illustrate the power of our model relative to the state of the art methods. | 0reject
|
Title: Towards Hierarchical Importance Attribution: Explaining Compositional Semantics for Neural Sequence Models. Abstract: The impressive performance of neural networks on natural language processing tasks attributes to their ability to model complicated word and phrase compositions. To explain how the model handles semantic compositions, we study hierarchical explanation of neural network predictions. We identify non-additivity and context independent importance attributions within hierarchies as two desirable properties for highlighting word and phrase compositions. We show some prior efforts on hierarchical explanations, e.g. contextual decomposition, do not satisfy the desired properties mathematically, leading to inconsistent explanation quality in different models. In this paper, we start by proposing a formal and general way to quantify the importance of each word and phrase. Following the formulation, we propose Sampling and Contextual Decomposition (SCD) algorithm and Sampling and Occlusion (SOC) algorithm. Human and metrics evaluation on both LSTM models and BERT Transformer models on multiple datasets show that our algorithms outperform prior hierarchical explanation algorithms. Our algorithms help to visualize semantic composition captured by models, extract classification rules and improve human trust of models. | 1accept
|
Title: Regularizing Black-box Models for Improved Interpretability. Abstract: Most of the work on interpretable machine learning has focused on designingeither inherently interpretable models, which typically trade-off accuracyfor interpretability, or post-hoc explanation systems, which lack guarantees about their explanation quality. We explore an alternative to theseapproaches by directly regularizing a black-box model for interpretabilityat training time. Our approach explicitly connects three key aspects ofinterpretable machine learning: (i) the model’s internal interpretability, (ii)the explanation system used at test time, and (iii) the metrics that measureexplanation quality. Our regularization results in substantial improvementin terms of the explanation fidelity and stability metrics across a range ofdatasets and black-box explanation systems while slightly improving accuracy. Finally, we justify theoretically that the benefits of our regularizationgeneralize to unseen points. | 0reject
|
Title: Joint Descent: Training and Tuning Simultaneously. Abstract: Typically in machine learning, training and tuning are done in an alternating manner: for a fixed set of hyperparameters $y$, we apply gradient descent to our objective $f(x, y)$ over trainable variables $x$ until convergence; then, we apply a tuning step over $y$ to find another promising setting of hyperparameters. Because the full training cycle is completed before a tuning step is applied, the optimization procedure greatly emphasizes the gradient step, which seems justified as first-order methods provides a faster convergence rate. In this paper, we argue that an equal emphasis on training and tuning lead to faster convergence both theoretically and empirically. We present Joint Descent (JD) and a novel theoretical analysis of acceleration via an unbiased gradient estimate to give an optimal iteration complexity of $O(\sqrt{\kappa}n_y\log(n/\epsilon))$, where $\kappa$ is the condition number and $n_y$ is the dimension of $y$. This provably improves upon the naive classical bound and implies that we essentially train for free if we apply equal emphasis on training and tuning steps. Empirically, we observe that an unbiased gradient estimate achieves the best convergence results, supporting our theory. | 2withdrawn
|
Title: Functional Bayesian Neural Networks for Model Uncertainty Quantification. Abstract: In this paper, we extend the Bayesian neural network to functional Bayesian neural network with functional Monte Carlo methods that use the samples of functionals instead of samples of networks' parameters for inference to overcome the curse of dimensionality for uncertainty quantification. Based on the previous work on Riemannian Langevin dynamics, we propose the stochastic gradient functional Riemannian dynamics for training functional Bayesian neural network. We show the effectiveness and efficiency of our proposed approach with various experiments. | 0reject
|
Title: Learning to Encode Text as Human-Readable Summaries using Generative Adversarial Networks. Abstract: Auto-encoders compress input data into a latent-space representation and reconstruct the original data from the representation. This latent representation is not easily interpreted by humans. In this paper, we propose training an auto-encoder that encodes input text into human-readable sentences. The auto-encoder is composed of a generator and a reconstructor. The generator encodes the input text into a shorter word sequence, and the reconstructor recovers the generator input from the generator output.
To make the generator output human-readable, a discriminator restricts the output of the generator to resemble human-written sentences. By taking the generator output as the summary of the input text, abstractive summarization is achieved without document-summary pairs as training data. Promising results are shown on both English and Chinese corpora. | 0reject
|
Title: Causal Scene BERT: Improving object detection by searching for challenging groups. Abstract: Autonomous vehicles (AV) rely on learning-based perception modules parametrized with neural networks for tasks like object detection. These modules frequently have low expected error overall but high error on atypical groups of data due to biases inherent in the training process. Multiple heuristics are employed to identify "failures" in AVs, a typical example being driver interventions. After identification, a human team combs through the associated data to group perception failures that share common causes. More data from these groups is then collected and annotated before retraining the model to fix the issue. In other words, error groups are found and addressed in hindsight as they appear. Our main contribution is a pseudo-automatic method to discover such groups in foresight by performing causal interventions on simulated driving scenes. To keep our interventions on the data manifold, we use masked language models. We verify that the prioritized groups found via intervention are challenging for the object detector and show that retraining with data collected from these groups helps inordinately compared to adding more IID data. We also release software to run interventions in simulated scenes, which we hope will benefit the causality community. | 2withdrawn
|
Title: Exploring Sentence Vectors Through Automatic Summarization. Abstract: Vector semantics, especially sentence vectors, have recently been used successfully in many areas of natural language processing. However, relatively little work has explored the internal structure and properties of spaces of sentence vectors. In this paper, we will explore the properties of sentence vectors by studying a particular real-world application: Automatic Summarization. In particular, we show that cosine similarity between sentence vectors and document vectors is strongly correlated with sentence importance and that vector semantics can identify and correct gaps between the sentences chosen so far and the document. In addition, we identify specific dimensions which are linked to effective summaries. To our knowledge, this is the first time specific dimensions of sentence embeddings have been connected to sentence properties. We also compare the features of different methods of sentence embeddings. Many of these insights have applications in uses of sentence embeddings far beyond summarization. | 0reject
|
Title: Data augmentation instead of explicit regularization. Abstract: Modern deep artificial neural networks have achieved impressive results through models with very large capacity---compared to the number of training examples---that control overfitting with the help of different forms of regularization. Regularization can be implicit, as is the case of stochastic gradient descent or parameter sharing in convolutional layers, or explicit. Most common explicit regularization techniques, such as dropout and weight decay, reduce the effective capacity of the model and typically require the use of deeper and wider architectures to compensate for the reduced capacity. Although these techniques have been proven successful in terms of results, they seem to waste capacity. In contrast, data augmentation techniques reduce the generalization error by increasing the number of training examples and without reducing the effective capacity. In this paper we systematically analyze the effect of data augmentation on some popular architectures and conclude that data augmentation alone---without any other explicit regularization techniques---can achieve the same performance or higher as regularized models, especially when training with fewer examples. | 0reject
|
Title: Source-Target Unified Knowledge Distillation for Memory-Efficient Federated Domain Adaptation on Edge Devices. Abstract: To support local inference on an edge device, it is necessary to deploy a compact machine learning model on such a device.
When such a compact model is applied to a new environment, its inference accuracy can be degraded if the target data from the new environment have a different distribution from the source data that are used for model training.
To ensure high inference accuracy in the new environment, it is indispensable to adapt the compact model to the target data.
However, to protect users' privacy, the target data cannot be sent to a centralized server for joint training with the source data. Furthermore, utilizing the target data to directly train the compact model cannot achieve sufficient inference accuracy due to its low model capacity.
To this end, a scheme called source-target unified knowledge distillation (STU-KD) is developed in this paper. It starts with a large pretrained model loaded onto the edge device, and then the knowledge of the large model is transferred to the compact model via knowledge distillation.
Since training the large model leads to large memory consumption, a domain adaptation method called lite-residual hypothesis transfer is designed to achieve memory-efficient adaptation to the target data on the edge device. Moreover, to prevent the compact model from forgetting the knowledge of the source data during knowledge distillation, a collaborative knowledge distillation (Co-KD) method is developed to unify the source data on the server and the target data on the edge device to train the compact model. STU-KD can be easily integrated with secure aggregation so that the server cannot obtain the true model parameters of the compact model. Extensive experiments conducted upon several objective recognition tasks show that STU-KD can improve the inference accuracy by up to $14.7\%$, as compared to the state-of-the-art schemes. Results also reveal that the inference accuracy of the compact model is not impacted by incorporating secure aggregation into STU-KD. | 0reject
|
Title: Going Deeper with Lean Point Networks. Abstract: We introduce three generic point cloud processing blocks that improve both accuracy and memory consumption of multiple state-of-the-art networks, thus allowing to design deeper and more accurate networks.
The novel processing blocks that facilitate efficient information flow are a convolution-type operation block for point sets that blends neighborhood information in a memory-efficient manner; a multi-resolution point cloud processing block; and a crosslink block that efficiently shares information across low- and high-resolution processing branches. Combining these blocks, we design significantly wider and deeper architectures.
We extensively evaluate the proposed architectures on multiple point segmentation benchmarks (ShapeNetPart, ScanNet, PartNet) and report systematic improvements in terms of both accuracy and memory consumption by using our generic modules in conjunction with multiple recent architectures (PointNet++, DGCNN, SpiderCNN, PointCNN). We report a 9.7% increase in IoU on the PartNet dataset, which is the most complex, while decreasing memory footprint by 57%. | 2withdrawn
|
Title: Boosting Ticket: Towards Practical Pruning for Adversarial Training with Lottery Ticket Hypothesis. Abstract: Recent research has proposed the lottery ticket hypothesis, suggesting that for a deep neural network, there exist trainable sub-networks performing equally or better than the original model with commensurate training steps. While this discovery is insightful, finding proper sub-networks requires iterative training and pruning. The high cost incurred limits the applications of the lottery ticket hypothesis. We show there exists a subset of the aforementioned sub-networks that converge significantly faster during the training process and thus can mitigate the cost issue. We conduct extensive experiments to show such sub-networks consistently exist across various model structures for a restrictive setting of hyperparameters (e.g., carefully selected learning rate, pruning ratio, and model capacity). As a practical application of our findings, we demonstrate that such sub-networks can help in cutting down the total time of adversarial training, a standard approach to improve robustness, by up to 49% on CIFAR-10 to achieve the state-of-the-art robustness. | 2withdrawn
|
Title: Text Infilling. Abstract: Recent years have seen remarkable progress of text generation in different contexts, including the most common setting of generating text from scratch, the increasingly popular paradigm of retrieval and editing, and others. Text infilling, which fills missing text portions of a sentence or paragraph, is also of numerous use in real life. Previous work has focused on restricted settings, by either assuming single word per missing portion, or limiting to single missing portion to the end of text. This paper studies the general task of text infilling, where the input text can have an arbitrary number of portions to be filled, each of which may require an arbitrary unknown number of tokens.
We develop a self-attention model with segment-aware position encoding for precise global context modeling.
We further create a variety of supervised data by masking out text in different domains with varying missing ratios and mask strategies. Extensive experiments show the proposed model performs significantly better than other methods, and generates meaningful text patches. | 0reject
|
Title: Double Neural Counterfactual Regret Minimization. Abstract: Counterfactual regret minimization (CRF) is a fundamental and effective technique for solving imperfect information games. However, the original CRF algorithm only works for discrete state and action spaces, and the resulting strategy is maintained as a tabular representation. Such tabular representation limits the method from being directly applied to large games and continuing to improve from a poor strategy profile. In this paper, we propose a double neural representation for the Imperfect Information Games, where one neural network represents the cumulative regret, and the other represents the average strategy. Furthermore, we adopt the counterfactual regret minimization algorithm to optimize this double neural representation. To make neural learning efficient, we also developed several novel techniques including a robust sampling method, mini-batch Monte Carlo counterfactual regret minimization (MCCFR) and Monte Carlo counterfactual regret minimization plus (MCCFR+) which may be of independent interests. Experimentally, we demonstrate that the proposed double neural algorithm converges significantly better than the reinforcement learning counterpart. | 0reject
|
Title: Learning a Latent Search Space for Routing Problems using Variational Autoencoders. Abstract: Methods for automatically learning to solve routing problems are rapidly improving in performance. While most of these methods excel at generating solutions quickly, they are unable to effectively utilize longer run times because they lack a sophisticated search component. We present a learning-based optimization approach that allows a guided search in the distribution of high-quality solutions for a problem instance. More precisely, our method uses a conditional variational autoencoder that learns to map points in a continuous (latent) search space to high-quality, instance-specific routing problem solutions. The learned space can then be searched by any unconstrained continuous optimization method. We show that even using a standard differential evolution search strategy our approach is able to outperform existing purely machine learning based approaches. | 1accept
|
Title: Provably Faster Algorithms for Bilevel Optimization and Applications to Meta-Learning. Abstract: Bilevel optimization has arisen as a powerful tool for many machine learning problems such as meta-learning, hyperparameter optimization, and reinforcement learning. In this paper, we investigate the nonconvex-strongly-convex bilevel optimization problem. For deterministic bilevel optimization, we provide a comprehensive finite-time convergence analysis for two popular algorithms respectively based on approximate implicit differentiation (AID) and iterative differentiation (ITD). For the AID-based method, we orderwisely improve the previous finite-time convergence analysis due to a more practical parameter selection as well as a warm start strategy, and for the ITD-based method we establish the first theoretical convergence rate. Our analysis also provides a quantitative comparison between ITD and AID based approaches. For stochastic bilevel optimization, we propose a novel algorithm named stocBiO, which features a sample-efficient hypergradient estimator using efficient Jacobian- and Hessian-vector product computations. We provide the finite-time convergence guarantee for stocBiO, and show that stocBiO outperforms the best known computational complexities orderwisely with respect to the condition number $\kappa$ and the target accuracy $\epsilon$. We further validate our theoretical results and demonstrate the efficiency of bilevel optimization algorithms by the experiments on meta-learning and hyperparameter optimization. | 0reject
|
Title: GRAPH NEIGHBORHOOD ATTENTIVE POOLING. Abstract: Network representation learning (NRL) is a powerful technique for learning low-dimensional vector representation of high-dimensional and sparse graphs. Most studies explore the structure and meta data associated with the graph using random walks and employ a unsupervised or semi-supervised learning schemes. Learning in these methods is context-free, because only a single representation per node is learned. Recently studies have argued on the sufficiency of a single representation and proposed a context-sensitive approach that proved to be highly effective in applications such as link prediction and ranking.
However, most of these methods rely on additional textual features that require RNNs or CNNs to capture high-level features or rely on a community detection algorithm to identifying multiple contexts of a node.
In this study, without requiring additional features nor a community detection algorithm, we propose a novel context-sensitive algorithm called GAP that learns to attend on different part of a node’s neighborhood using attentive pooling networks. We show the efficacy of GAP using three real-world datasets on link prediction and node clustering tasks and compare it against 10 popular and state-of-the-art (SOTA) baselines. GAP consistently outperforms them and achieves up to ≈9% and ≈20% gain over the best performing methods on link prediction and clustering tasks, respectively. | 0reject
|
Title: Clearing the Path for Truly Semantic Representation Learning. Abstract: The performance of $\beta$-Variational-Autoencoders ($\beta$-VAEs) and their variants on learning semantically meaningful, disentangled representations is unparalleled. On the other hand, there are theoretical arguments suggesting impossibility of unsupervised disentanglement. In this work, we show that small perturbations of existing datasets hide the convenient correlation structure that is easily exploited by VAE-based architectures. To demonstrate this, we construct modified versions of the standard datasets on which (i) the generative factors are perfectly preserved; (ii) each image undergoes a transformation barely visible to the human eye; (iii) the leading disentanglement architectures fail to produce disentangled representations. We intend for these datasets to play a role in separating correlation-based models from those that discover the true causal structure.
The construction of the modifications is non-trivial and relies on recent progress on mechanistic understanding of $\beta$-VAEs and their connection to PCA, while also providing additional insights that might be of stand-alone interest. | 0reject
|
Title: INFERENCE, PREDICTION, AND ENTROPY RATE OF CONTINUOUS-TIME, DISCRETE-EVENT PROCESSES. Abstract: The inference of models, prediction of future symbols, and entropy rate estimation of discrete-time, discrete-event processes is well-worn ground. However, many time series are better conceptualized as continuous-time, discrete-event processes. Here, we provide new methods for inferring models, predicting future symbols, and estimating the entropy rate of continuous-time, discrete-event processes. The methods rely on an extension of Bayesian structural inference that takes advantage of neural network’s universal approximation power. Based on experiments with simple synthetic data, these new methods seem to be competitive with state-of- the-art methods for prediction and entropy rate estimation as long as the correct model is inferred. | 0reject
|
Title: Learning Curves for Analysis of Deep Networks. Abstract: A learning curve models a classifier’s test error as a function of the number of training samples. Prior works show that learning curves can be used to select model parameters and extrapolate performance. We investigate how to use learning curves to analyze the impact of design choices, such as pre-training, architecture, and data augmentation. We propose a method to robustly estimate learning curves, abstract their parameters into error and data-reliance, and evaluate the effectiveness of different parameterizations. We also provide several interesting observations based on learning curves for a variety of image classification models. | 0reject
|
Title: Help Me Explore: Minimal Social Interventions for Graph-Based Autotelic Agents. Abstract: In the quest for autonomous agents learning open-ended repertoires of skills, most works take a Piagetian perspective: learning trajectories are the results of interactions between developmental agents and their physical environment. The Vygotskian perspective, on the other hand, emphasizes the centrality of the socio-cultural environment: higher cognitive functions emerge from transmissions of socio-cultural processes internalized by the agent. This paper acknowledges these two perspectives and presents GANGSTR, a hybrid agent engaging in both individual and social goal-directed exploration. In a 5-block manipulation domain, GANGSTR discovers and learns to master tens of thousands of configurations. In individual phases, the agent engages in autotelic learning; it generates, pursues and makes progress towards its own goals. To this end, it builds a graph to represent the network of discovered configuration and to navigate between them. In social phases, a simulated social partner suggests goal configurations at the frontier of the agent’s current capabilities. This paper makes two contributions: 1) a minimal social interaction protocol called Help Me Explore (HME); 2) GANGSTR, a graph-based autotelic agent. As this paper shows, coupling individual and social exploration enables the GANGSTR agent to discover and master the most complex configurations (e.g. stacks of 5 blocks) with only minimal intervention. | 0reject
|
Title: CoPhy: Counterfactual Learning of Physical Dynamics. Abstract: Understanding causes and effects in mechanical systems is an essential component of reasoning in the physical world. This work poses a new problem of counterfactual learning of object mechanics from visual input. We develop the CoPhy benchmark to assess the capacity of the state-of-the-art models for causal physical reasoning in a synthetic 3D environment and propose a model for learning the physical dynamics in a counterfactual setting. Having observed a mechanical experiment that involves, for example, a falling tower of blocks, a set of bouncing balls or colliding objects, we learn to predict how its outcome is affected by an arbitrary intervention on its initial conditions, such as displacing one of the objects in the scene. The alternative future is predicted given the altered past and a latent representation of the confounders learned by the model in an end-to-end fashion with no supervision. We compare against feedforward video prediction baselines and show how observing alternative experiences allows the network to capture latent physical properties of the environment, which results in significantly more accurate predictions at the level of super human performance. | 1accept
|
Title: Double Generative Adversarial Networks for Conditional Independence Testing. Abstract: In this article, we consider the problem of high-dimensional conditional independence testing, which is a key building block in statistics and machine learning. We propose a double generative adversarial networks (GAN)-based inference procedure. We first introduce a double GANs framework to learn two generators, and integrate the two generators to construct a doubly-robust test statistic. We next consider multiple generalized covariance measures, and take their maximum as our test statistic. Finally, we obtain the empirical distribution of our test statistic through multiplier bootstrap. We show that our test controls type-I error, while the
power approaches one asymptotically. More importantly, these theoretical guarantees are obtained under much weaker and practically more feasible conditions compared to existing tests. We demonstrate the efficacy of our test through both synthetic and real datasets. | 0reject
|
Title: Conditional Network Embeddings. Abstract: Network Embeddings (NEs) map the nodes of a given network into $d$-dimensional Euclidean space $\mathbb{R}^d$. Ideally, this mapping is such that 'similar' nodes are mapped onto nearby points, such that the NE can be used for purposes such as link prediction (if 'similar' means being 'more likely to be connected') or classification (if 'similar' means 'being more likely to have the same label'). In recent years various methods for NE have been introduced, all following a similar strategy: defining a notion of similarity between nodes (typically some distance measure within the network), a distance measure in the embedding space, and a loss function that penalizes large distances for similar nodes and small distances for dissimilar nodes.
A difficulty faced by existing methods is that certain networks are fundamentally hard to embed due to their structural properties: (approximate) multipartiteness, certain degree distributions, assortativity, etc. To overcome this, we introduce a conceptual innovation to the NE literature and propose to create \emph{Conditional Network Embeddings} (CNEs); embeddings that maximally add information with respect to given structural properties (e.g. node degrees, block densities, etc.). We use a simple Bayesian approach to achieve this, and propose a block stochastic gradient descent algorithm for fitting it efficiently.
We demonstrate that CNEs are superior for link prediction and multi-label classification when compared to state-of-the-art methods, and this without adding significant mathematical or computational complexity. Finally, we illustrate the potential of CNE for network visualization. | 1accept
|
Title: iPrune: A Magnitude Based Unstructured Pruning Method for Efficient Binary Networks in Hardware. Abstract: Modern image recognition models span millions of parameters occupying several megabytes and sometimes gigabytes of space, making it difficult to run on resource constrained edge hardware. Binary Neural Networks address this problem by reducing the memory requirements (one single bit per weight and/or activation). The computation requirement and power consumption are also reduced accordingly. Nevertheless, each neuron in such networks has a large number of inputs, making it difficult to implement them efficiently in binary hardware accelerators, especially LUT-based approaches.
In this work, we present a pruning algorithm and associated results on convolutional and dense layers from aforementioned binary networks. We reduce the computation by 4-70x and the memory by 190-2200x with less than 2% loss of accuracy on MNIST and less than 3% loss of accuracy on CIFAR-10 compared to full precision, fully connected equivalents. Compared to very recent work on pruning for binary networks, we still have a gain of 1% on the precision and up to 30% reduction in memory (526KiB vs 750KiB). | 0reject
|
Title: On the Selection of Initialization and Activation Function for Deep Neural Networks. Abstract: The weight initialization and the activation function of deep neural networks have a crucial impact on the performance of the training procedure. An inappropriate selection can lead to the loss of information of the input during forward propagation and the exponential vanishing/exploding of gradients during back-propagation. Understanding the theoretical properties of untrained random networks is key to identifying which deep networks may be trained successfully as recently demonstrated by Schoenholz et al. (2017) who showed that for deep feedforward neural networks only a specific choice of hyperparameters known as the `edge of chaos' can lead to good performance.
We complete this analysis by providing quantitative results showing that, for a class of ReLU-like activation functions, the information propagates indeed deeper for an initialization at the edge of chaos. By further extending this analysis, we identify a class of activation functions that improve the information propagation over ReLU-like functions. This class includes the Swish activation, $\phi_{swish}(x) = x \cdot \text{sigmoid}(x)$, used in Hendrycks & Gimpel (2016),
Elfwing et al. (2017) and Ramachandran et al. (2017). This provides a theoretical grounding for the excellent empirical performance of $\phi_{swish}$ observed in these contributions. We complement those previous results by illustrating the benefit of using a random initialization on the edge of chaos in this context. | 0reject
|
Title: A Mutual Information Maximization Perspective of Language Representation Learning. Abstract: We show state-of-the-art word representation learning methods maximize an objective function that is a lower bound on the mutual information between different parts of a word sequence (i.e., a sentence). Our formulation provides an alternative perspective that unifies classical word embedding models (e.g., Skip-gram) and modern contextual embeddings (e.g., BERT, XLNet). In addition to enhancing our theoretical understanding of these methods, our derivation leads to a principled framework that can be used to construct new self-supervised tasks. We provide an example by drawing inspirations from related methods based on mutual information maximization that have been successful in computer vision, and introduce a simple self-supervised objective that maximizes the mutual information between a global sentence representation and n-grams in the sentence. Our analysis offers a holistic view of representation learning methods to transfer knowledge and translate progress across multiple domains (e.g., natural language processing, computer vision, audio processing). | 1accept
|
Title: How the Softmax Activation Hinders the Detection of Adversarial and Out-of-Distribution Examples in Neural Networks. Abstract: Despite having excellent performances for a wide variety of tasks, modern neural networks are unable to provide a prediction with a reliable confidence estimate which would allow to detect misclassifications. This limitation is at the heart of what is known as an adversarial example, where the network provides a wrong prediction associated with a strong confidence to a slightly modified image. Moreover, this overconfidence issue has also been observed for out-of-distribution data. We show through several experiments that the softmax activation, usually placed as the last layer of modern neural networks, is partly responsible for this behaviour. We give qualitative insights about its impact on the MNIST dataset, showing that relevant information present in the logits is lost once the softmax function is applied. The same observation is made through quantitative analysis, as we show that two out-of-distribution and adversarial example detectors obtain competitive results when using logit values as inputs, but provide considerably lower performances if they use softmax probabilities instead: from 98.0% average AUROC to 56.8% in some settings. These results provide evidence that the softmax activation hinders the detection of adversarial and out-of-distribution examples, as it masks a significant part of the relevant information present in the logits. | 0reject
|
Title: Test-Time Adaptation and Adversarial Robustness. Abstract: This paper studies test-time adaptation in the context of adversarial robustness. We formulate an adversarial threat model for test-time adaptation, where the defender may have a unique advantage as the adversarial game becomes a maximin game, instead of a minimax game as in the classic adversarial robustness threat model. We then study whether the maximin threat model admits more ``good solutions'' than the minimax threat model, and is thus \emph{strictly weaker}. For this purpose, we first present a provable separation between the two threat models in a natural Gaussian data model. For deep learning, while we do not have a proof, we propose a candidate, Domain Adversarial Neural Networks (${\sf DANN}$), an algorithm designed for unsupervised domain adaptation, by showing that it provides nontrivial robustness in the test-time maximin threat model against strong transfer attacks and adaptive attacks. This is somewhat surprising since ${\sf DANN}$ is not designed specifically for adversarial robustness (e.g., against norm-based attacks), and provides no robustness in the minimax model. Complementing these results, we show that recent data-oblivious test-time adaptations can be easily attacked even with simple transfer attacks. We conclude the paper with various future directions of studying adversarially robust test-time adaptation. | 0reject
|
Title: Lagrangian Fluid Simulation with Continuous Convolutions. Abstract: We present an approach to Lagrangian fluid simulation with a new type of convolutional network. Our networks process sets of moving particles, which describe fluids in space and time. Unlike previous approaches, we do not build an explicit graph structure to connect the particles but use spatial convolutions as the main differentiable operation that relates particles to their neighbors. To this end we present a simple, novel, and effective extension of N-D convolutions to the continuous domain. We show that our network architecture can simulate different materials, generalizes to arbitrary collision geometries, and can be used for inverse problems. In addition, we demonstrate that our continuous convolutions outperform prior formulations in terms of accuracy and speed.
| 1accept
|
Title: RNNs Incrementally Evolving on an Equilibrium Manifold: A Panacea for Vanishing and Exploding Gradients?. Abstract: Recurrent neural networks (RNNs) are particularly well-suited for modeling long-term dependencies in sequential data, but are notoriously hard to train because the error backpropagated in time either vanishes or explodes at an exponential rate. While a number of works attempt to mitigate this effect through gated recurrent units, skip-connections, parametric constraints and design choices, we propose a novel incremental RNN (iRNN), where hidden state vectors keep track of incremental changes, and as such approximate state-vector increments of Rosenblatt's (1962) continuous-time RNNs. iRNN exhibits identity gradients and is able to account for long-term dependencies (LTD). We show that our method is computationally efficient overcoming overheads of many existing methods that attempt to improve RNN training, while suffering no performance degradation. We demonstrate the utility of our approach with extensive experiments and show competitive performance against standard LSTMs on LTD and other non-LTD tasks.
| 1accept
|
Title: Variance Reduction in Hierarchical Variational Autoencoders. Abstract: Variational autoencoders with deep hierarchies of stochastic layers have been known to suffer from the problem of posterior collapse, where the top layers fall back to the prior and become independent of input.
We suggest that the hierarchical VAE objective explicitly includes the variance of the function parameterizing the mean and variance of the latent Gaussian distribution which itself is often a high variance function.
Building on this we generalize VAE neural networks by incorporating a smoothing parameter motivated by Gaussian analysis to reduce variance in parameterizing functions and show that this can help to solve the problem of posterior collapse.
We further show that under such smoothing the VAE loss exhibits a phase transition, where the top layer KL divergence sharply drops to zero at a critical value of the smoothing parameter.
We validate the phenomenon across model configurations and datasets. | 0reject
|
Title: Clustering-friendly Representation Learning via Instance Discrimination and Feature Decorrelation. Abstract: Clustering is one of the most fundamental tasks in machine learning. Recently, deep clustering has become a major trend in clustering techniques. Representation learning often plays an important role in the effectiveness of deep clustering, and thus can be a principal cause of performance degradation. In this paper, we propose a clustering-friendly representation learning method using instance discrimination and feature decorrelation. Our deep-learning-based representation learning method is motivated by the properties of classical spectral clustering. Instance discrimination learns similarities among data and feature decorrelation removes redundant correlation among features. We utilize an instance discrimination method in which learning individual instance classes leads to learning similarity among instances. Through detailed experiments and examination, we show that the approach can be adapted to learning a latent space for clustering. We design novel softmax-formulated decorrelation constraints for learning. In evaluations of image clustering using CIFAR-10 and ImageNet-10, our method achieves accuracy of 81.5% and 95.4%, respectively. We also show that the softmax-formulated constraints are compatible with various neural networks. | 1accept
|
Title: Disentangling Adversarial Robustness in Directions of the Data Manifold. Abstract: Using generative models (GAN or VAE) to craft adversarial examples, i.e. generative adversarial examples, has received increasing attention in recent years. Previous studies showed that the generative adversarial examples work differently compared to that of the regular adversarial examples in many aspects, such as attack rates, perceptibility, and generalization. But the reasons causing the differences between regular and generative adversarial examples are unclear. In this work, we study the theoretical properties of the attacking mechanisms of the two kinds of adversarial examples in the Gaussian mixture data model case. We prove that adversarial robustness can be disentangled in directions of the data manifold. Specifically, we find that: 1. Regular adversarial examples attack in directions of small variance of the data manifold, while generative adversarial examples attack in directions of large variance. 2. Standard adversarial training increases model robustness by extending the data manifold boundary in directions of small variance, while on the contrary, adversarial training with generative adversarial examples increases model robustness by extending the data manifold boundary directions of large variance. In experiments, we demonstrate that these phenomena also exist on real datasets. Finally, we study the robustness trade-off between generative and regular adversarial examples. We show that the conflict between regular and generative adversarial examples is much smaller than the conflict between regular adversarial examples of different norms. | 0reject
|
Title: GeDi: Generative Discriminator Guided Sequence Generation. Abstract: While large-scale language models (LMs) are able to imitate the distribution of natural language well enough to generate realistic text, it is difficult to control which regions of the distribution they generate. This is especially problematic because datasets used for training large LMs usually contain significant toxicity, hate, bias, and negativity. We propose GeDi as an efficient method for using smaller LMs as generative discriminators to guide generation from large LMs to make them safer and more controllable. GeDi guides generation at each step by computing classification probabilities for all possible next tokens via Bayes rule by normalizing over two class-conditional distributions; one conditioned on the desired attribute, or control code, and another conditioned on the undesired attribute, or anti control code. We find that GeDi gives controllability on par with or better than the state of the art method in a variety of settings, while also achieving generation speeds more than $30$ times faster. Additionally, training GeDi on only three topics allows us to controllably generate new topics zero-shot from just a keyword. Lastly, we show that GeDi can make GPT-2 and GPT-3 significantly less toxic without sacrificing on linguistic fluency, making it by far the most practical existing method for detoxifying large language models while maintaining a fast generation speed. | 0reject
|
Title: Neural MMO: A massively multiplayer game environment for intelligent agents. Abstract: We present an artificial intelligence research platform inspired by the human game genre of MMORPGs (Massively Multiplayer Online Role-Playing Games, a.k.a. MMOs). We demonstrate how this platform can be used to study behavior and learning in large populations of neural agents. Unlike currently popular game environments, our platform supports persistent environments, with variable number of agents, and open-ended task descriptions. The emergence of complex life on Earth is often attributed to the arms race that ensued from a huge number of organisms all competing for finite resources. Our platform aims to simulate this setting in microcosm: we conduct a series of experiments to test how large-scale multiagent competition can incentivize the development of skillful behavior. We find that population size magnifies the complexity of the behaviors that emerge and results in agents that out-compete agents trained in smaller populations. | 0reject
|
Title: Contextual HyperNetworks for Novel Feature Adaptation. Abstract: While deep learning has obtained state-of-the-art results in many applications, the adaptation of neural network architectures to incorporate new features remains a research challenge. This issue is particularly severe in online learning settings, where new features are added continually with few or no associated observations. As such, methods for adapting neural networks to novel features which are both time and data-efficient are desired. To address this, we propose the Contextual HyperNetwork (CHN), which predicts the network weights associated with new features by incorporating information from both existing data as well as the few observations for the new feature and any associated feature metadata. At prediction time, the CHN requires only a single forward pass through a small neural network, yielding a significant speed-up when compared to re-training and fine-tuning approaches. In order to showcase the performance of CHNs, in this work we use a CHN to augment a partial variational autoencoder (P-VAE), a flexible deep generative model which can impute the values of missing features in sparsely-observed data. We show that this system obtains significantly improved performance for novel feature adaptation over existing imputation and meta-learning baselines across recommender systems, e-learning, and healthcare tasks. | 0reject
|
Title: A Generalised Inverse Reinforcement Learning Framework. Abstract: The global objective of inverse Reinforcement Learning (IRL) is to estimate the unknown cost function of some MDP based on observed trajectories generated by (approximate) optimal policies. The classical approach consists in tuning this cost function so that associated optimal trajectories (that minimise the cumulative discounted cost, i.e. the classical RL loss) are “similar” to the observed ones. Prior contributions focused on penalising degenerate solutions and improving algorithmic scalability. Quite orthogonally to them, we question the pertinence of characterising optimality with respect to the cumulative discounted cost as it induces an implicit bias against policies with longer mixing times. State of the art value based RL algorithms circumvent this issue by solving for the fixed point of the Bellman optimality operator, a stronger criterion that is not well defined for the inverse problem.
To alleviate this bias in IRL, we introduce an alternative training loss that puts more weights on future states which yields a reformulation of the (maximum entropy) IRL problem. The algorithms we devised exhibit enhanced performances (and similar tractability) than off-the-shelf ones in multiple OpenAI gym environments. | 0reject
|
Title: cosFormer: Rethinking Softmax In Attention. Abstract: Transformer has shown great successes in natural language processing, computer vision, and audio processing. As one of its core components, the softmax attention helps to capture long-range dependencies yet prohibits its scale-up due to the quadratic space and time complexity to the sequence length. Kernel methods are often adopted to reduce the complexity by approximating the softmax operator. Nevertheless, due to the approximation errors, their performances vary in different tasks/corpus and suffer crucial performance drops when compared with the vanilla softmax attention. In this paper, we propose a linear transformer called cosFormer that can achieve comparable or better accuracy to the vanilla transformer in both casual and cross attentions. cosFormer is based on two key properties of softmax attention: i). non-negativeness of the attention matrix; ii). a non-linear re-weighting scheme that can concentrate the distribution of the attention matrix. As its linear substitute, cosFormer fulfills these properties with a linear operator and a cosine-based distance re-weighting mechanism. Extensive experiments on language modeling and text understanding tasks demonstrate the effectiveness of our method. We further examine our method on long sequences and achieve state-of-the-art performance on the Long-Range Arena benchmark. The source code is available at https://github.com/OpenNLPLab/cosFormer. | 1accept
|
Title: SNODE: Spectral Discretization of Neural ODEs for System Identification. Abstract: This paper proposes the use of spectral element methods \citep{canuto_spectral_1988} for fast and accurate training of Neural Ordinary Differential Equations (ODE-Nets; \citealp{Chen2018NeuralOD}) for system identification. This is achieved by expressing their dynamics as a truncated series of Legendre polynomials. The series coefficients, as well as the network weights, are computed by minimizing the weighted sum of the loss function and the violation of the ODE-Net dynamics. The problem is solved by coordinate descent that alternately minimizes, with respect to the coefficients and the weights, two unconstrained sub-problems using standard backpropagation and gradient methods. The resulting optimization scheme is fully time-parallel and results in a low memory footprint. Experimental comparison to standard methods, such as backpropagation through explicit solvers and the adjoint technique \citep{Chen2018NeuralOD}, on training surrogate models of small and medium-scale dynamical systems shows that it is at least one order of magnitude faster at reaching a comparable value of the loss function. The corresponding testing MSE is one order of magnitude smaller as well, suggesting generalization capabilities increase. | 1accept
|
Title: Automatically Learning Feature Crossing from Model Interpretation for Tabular Data. Abstract: Automatically feature generation is a major topic of automated machine learning. Among various feature generation approaches, feature crossing, which takes cross-product of sparse features, is a promising way to effectively capture the interactions among categorical features in tabular data. Previous works on feature crossing try to search in the set of all the possible cross feature fields. This is obviously not efficient when the size of original feature fields is large. Meanwhile, some deep learning-based methods combines deep neural networks and various interaction components. However, due to the existing of Deep Neural Networks (DNN), only a few cross features can be explicitly generated by the interaction components. Recently, piece-wise interpretation of DNN has been widely studied, and the piece-wise interpretations are usually inconsistent in different samples. Inspired by this, we give a definition of interpretation inconsistency in DNN, and propose a novel method called CrossGO, which selects useful cross features according to the interpretation inconsistency. The whole process of learning feature crossing can be done via simply training a DNN model and a logistic regression (LR) model. CrossGO can generate compact candidate set of cross feature fields, and promote the efficiency of searching. Extensive experiments have been conducted on several real-world datasets. Cross features generated by CrossGO can empower a simple LR model achieving approximate or even better performances comparing with complex DNN models. | 0reject
|
Title: Depth Completion using Plane-Residual Representation. Abstract: The basic framework of depth completion is to predict a pixel-wise dense depth map using very sparse input data.
In this paper, we try to solve this problem in a more effective way, by reformulating the regression-based depth estimation problem into a combination of depth plane classification and residual regression.
Our proposed approach is to initially densify sparse depth information by figuring out which plane a pixel should lie among a number of discretized depth planes, and then calculate the final depth value by predicting the distance from the specified plane.
This will help the network to lessen the burden of directly regressing the absolute depth information from none, and to effectively obtain more accurate depth prediction result with less computation power and inference time.
To do so, we firstly introduce a novel way of interpreting depth information with the closest depth plane label $p$ and a residual value $r$, as we call it, Plane-Residual (PR) representation.
We also propose a depth completion network utilizing PR representation consisting of a shared encoder and two decoders, where one classifies the pixel's depth plane label and the other one regresses the normalized distance from the classified depth plane.
By interpreting depth information in PR representation and using our corresponding depth completion network, we were able to acquire improved depth completion performance with faster computation, comparing to other recent approaches. | 2withdrawn
|
Title: Does Adversarial Transferability Indicate Knowledge Transferability?. Abstract: Despite the immense success that deep neural networks (DNNs) have achieved, \emph{adversarial examples}, which are perturbed inputs that aim to mislead DNNs to make mistakes, have recently led to great concerns. On the other hand, adversarial examples exhibit interesting phenomena, such as \emph{adversarial transferability}. DNNs also exhibit knowledge transfer, which is critical to improving learning efficiency and learning in domains that lack high-quality training data. To uncover the fundamental connections between these phenomena, we investigate and give an affirmative answer to the question: \emph{does adversarial transferability indicate knowledge transferability?} We theoretically analyze the relationship between adversarial transferability and knowledge transferability, and outline easily checkable sufficient conditions that identify when adversarial transferability indicates knowledge transferability. In particular, we show that composition with an affine function is sufficient to reduce the difference between two models when they possess high adversarial transferability. Furthermore, we provide empirical evaluation for different transfer learning scenarios on diverse datasets, showing a strong positive correlation between the adversarial transferability and knowledge transferability, thus illustrating that our theoretical insights are predictive of practice. | 0reject
|
Title: Visual Representation Learning Does Not Generalize Strongly Within the Same Domain. Abstract: An important component for generalization in machine learning is to uncover underlying latent factors of variation as well as the mechanism through which each factor acts in the world.
In this paper, we test whether 17 unsupervised, weakly supervised, and fully supervised representation learning approaches correctly infer the generative factors of variation in simple datasets (dSprites, Shapes3D, MPI3D) from controlled environments, and on our contributed CelebGlow dataset.
In contrast to prior robustness work that introduces novel factors of variation during test time, such as blur or other (un)structured noise, we here recompose, interpolate, or extrapolate only existing factors of variation from the training data set (e.g., small and medium-sized objects during training and large objects during testing). Models that learn the correct mechanism should be able to generalize to this benchmark.
In total, we train and test 2000+ models and observe that all of them struggle to learn the underlying mechanism regardless of supervision signal and architectural bias. Moreover, the generalization capabilities of all tested models drop significantly as we move from artificial datasets towards more realistic real-world datasets.
Despite their inability to identify the correct mechanism, the models are quite modular as their ability to infer other in-distribution factors remains fairly stable, providing only a single factor is out-of-distribution. These results point to an important yet understudied problem of learning mechanistic models of observations that can facilitate generalization. | 1accept
|
Title: Self-Monitoring Navigation Agent via Auxiliary Progress Estimation. Abstract: The Vision-and-Language Navigation (VLN) task entails an agent following navigational instruction in photo-realistic unknown environments. This challenging task demands that the agent be aware of which instruction was completed, which instruction is needed next, which way to go, and its navigation progress towards the goal. In this paper, we introduce a self-monitoring agent with two complementary components: (1) visual-textual co-grounding module to locate the instruction completed in the past, the instruction required for the next action, and the next moving direction from surrounding images and (2) progress monitor to ensure the grounded instruction correctly reflects the navigation progress. We test our self-monitoring agent on a standard benchmark and analyze our proposed approach through a series of ablation studies that elucidate the contributions of the primary components. Using our proposed method, we set the new state of the art by a significant margin (8% absolute increase in success rate on the unseen test set). Code is available at https://github.com/chihyaoma/selfmonitoring-agent. | 1accept
|
Title: Continuous Meta-Learning without Tasks. Abstract: Meta-learning is a promising strategy for learning to efficiently learn within new tasks, using data gathered from a distribution of tasks. However, the meta-learning literature thus far has focused on the task segmented setting, where at train-time, offline data is assumed to be split according to the underlying task, and at test-time, the algorithms are optimized to learn in a single task. In this work, we enable the application of generic meta-learning algorithms to settings where this task segmentation is unavailable, such as continual online learning with a time-varying task. We present meta-learning via online changepoint analysis (MOCA), an approach which augments a meta-learning algorithm with a differentiable Bayesian changepoint detection scheme. The framework allows both training and testing directly on time series data without segmenting it into discrete tasks. We demonstrate the utility of this approach on a nonlinear meta-regression benchmark as well as two meta-image-classification benchmarks. | 0reject
|
Title: Learning to Control Latent Representations for Few-Shot Learning of Named Entities. Abstract: Humans excel in continuously learning with small data without forgetting how to solve old problems.
However, neural networks require large datasets to compute latent representations across different tasks while minimizing a loss function. For example, a natural language understanding (NLU) system will often deal with emerging entities during its deployment as interactions with users in realistic scenarios will generate new and infrequent names, events, and locations. Here, we address this scenario by introducing a RL trainable controller that disentangles the representation learning of a neural encoder from its memory management role.
Our proposed solution is straightforward and simple: we train a controller to execute an optimal sequence of read and write operations on an external memory with the goal of leveraging diverse activations from the past and provide accurate predictions. Our approach is named Learning to Control (LTC) and allows few-shot learning with two degrees of memory plasticity. We experimentally show that our system obtains accurate results for few-shot learning of entity recognition in the Stanford Task-Oriented Dialogue dataset. | 0reject
|
Title: Learning a Max-Margin Classifier for Cross-Domain Sentiment Analysis. Abstract: Sentiment analysis is a costly yet necessary task for enterprises to study the opinions of their costumers to improve their products and services and to determine optimal marketing strategies. Due to existence of a wide range of domains across different products and services, cross-domain sentiment analysis methods have received significant attention in recent years. These methods mitigate the domain gap between different applications by training cross-domain generalizable classifiers which help to relax the need for individual data annotation per each domain. Most existing methods focus on learning domain-agnostic representations that are invariant with respect to both the source and the target domains. As a result, a classifier that is trained using annotated data in a source domain, would generalize well in a related target domain. In this work, we introduce a new domain adaptation method which induces large margins between different classes in an embedding space based on the notion of prototypical distribution. This embedding space is trained to be domain-agnostic by matching the data distributions across the domains. Large margins in the source domain help to reduce the effect of ``domain shift'' on the performance of a trained classifier in the target domain. Theoreticaland empirical analysis are provided to demonstrate that the method is effective. | 0reject
|
Title: Identifying Treatment Effects under Unobserved Confounding by Causal Representation Learning. Abstract: As an important problem of causal inference, we discuss the estimation of treatment effects under the existence of unobserved confounding. By representing the confounder as a latent variable, we propose Counterfactual VAE, a new variant of variational autoencoder, based on recent advances in identifiability of representation learning. Combining the identifiability and classical identification results of causal inference, under mild assumptions on the generative model and with small noise on the outcome, we theoretically show that the confounder is identifiable up to an affine transformation and then the treatment effects can be identified. Experiments on synthetic and semi-synthetic datasets demonstrate that our method matches the state-of-the-art, even under settings violating our formal assumptions. | 0reject
|
Title: On Position Embeddings in BERT. Abstract: Various Position Embeddings (PEs) have been proposed in Transformer based architectures~(e.g. BERT) to model word order. These are empirically-driven and perform well, but no formal framework exists to systematically study them. To address this, we present three properties of PEs that capture word distance in vector space: translation invariance, monotonicity, and symmetry. These properties formally capture the behaviour of PEs and allow us to reinterpret sinusoidal PEs in a principled way.
Moreover, we propose a new probing test (called `identical word probing') and mathematical indicators to quantitatively detect the general attention patterns with respect to the above properties. An empirical evaluation of seven PEs (and their combinations) for classification (GLUE) and span prediction (SQuAD) shows that: (1) both classification and span prediction benefit from translation invariance and local monotonicity, while symmetry slightly decreases performance;
(2) The fully-learnable absolute PE performs better in classification, while relative PEs perform better in span prediction. We contribute the first formal and quantitative analysis of desiderata for PEs, and a principled discussion about their correlation to the performance of typical downstream tasks. | 1accept
|
Title: S-System, Geometry, Learning, and Optimization: A Theory of Neural Networks. Abstract: We present a formal measure-theoretical theory of neural networks (NN) built on {\it probability coupling theory}. Particularly, we present an algorithm framework, Hierarchical Measure Group and Approximate System (HMGAS), nicknamed S-System, of which NNs are special cases. In addition to many other results, the framework enables us to prove that 1) NNs implement {\it renormalization group (RG)} using information geometry, which points out that the large scale property to renormalize is dual Bregman divergence and completes the analog between NNs and RG; 2) and under a set of {\it realistic} boundedness and diversity conditions, for {\it large size nonlinear deep} NNs with a class of losses, including the hinge loss, all local minima are global minima with zero loss errors, using random matrix theory. | 0reject
|
Title: Discriminative Variational Autoencoder for Continual Learning with Generative Replay. Abstract: Generative replay (GR) is a method to alleviate catastrophic forgetting in continual learning (CL) by generating previous task data and learning them together with the data from new tasks. In this paper, we propose discriminative variational autoencoder (DiVA) to address the GR-based CL problem. DiVA has class-wise discriminative latent embeddings by maximizing the mutual information between classes and latent variables of VAE. Thus, DiVA is directly applicable to classification and class-conditional generation which are efficient and effective properties in the GR-based CL scenario. Furthermore, we use a novel trick based on domain translation to cover natural images which is challenging to GR-based methods. As a result, DiVA achieved the competitive or higher accuracy compared to state-of-the-art algorithms in Permuted MNIST, Split MNIST, and Split CIFAR10 settings. | 0reject
|
Title: RvS: What is Essential for Offline RL via Supervised Learning?. Abstract: Recent work has shown that supervised learning alone, without temporal difference (TD) learning, can be remarkably effective for offline RL. When does this hold true, and which algorithmic components are necessary? Through extensive experiments, we boil supervised learning for offline RL down to its essential elements. In every environment suite we consider, simply maximizing likelihood with a two-layer feedforward MLP is competitive with state-of-the-art results of substantially more complex methods based on TD learning or sequence modeling with Transformers. Carefully choosing model capacity (e.g., via regularization or architecture) and choosing which information to condition on (e.g., goals or rewards) are critical for performance. These insights serve as a field guide for practitioners doing Reinforcement Learning via Supervised Learning (which we coin RvS learning). They also probe the limits of existing RvS methods, which are comparatively weak on random data, and suggest a number of open problems. | 1accept
|
Title: Deep processing of structured data. Abstract: We construct a general unified framework for learning representation of structured
data, i.e. data which cannot be represented as the fixed-length vectors (e.g. sets,
graphs, texts or images of varying sizes). The key factor is played by an intermediate
network called SAN (Set Aggregating Network), which maps a structured
object to a fixed length vector in a high dimensional latent space. Our main theoretical
result shows that for sufficiently large dimension of the latent space, SAN is
capable of learning a unique representation for every input example. Experiments
demonstrate that replacing pooling operation by SAN in convolutional networks
leads to better results in classifying images with different sizes. Moreover, its direct
application to text and graph data allows to obtain results close to SOTA, by
simpler networks with smaller number of parameters than competitive models. | 0reject
|
Title: Reflective Decoding: Unsupervised Paraphrasing and Abductive Reasoning. Abstract: Pretrained Language Models (LMs) generate text with remarkable quality, novelty, and coherence while semantically conditioning on context. Yet applying LMs to the seemingly simpler problems of paraphrasing and infilling currently requires supervision, since these tasks break the left-to-right generation setup of pretrained LMs. We present Reflective Decoding, a novel unsupervised approach to apply the capabilities of pretrained LMs to non-sequential tasks. Our approach is general and applicable to two distant tasks -- paraphrasing and abductive reasoning. It requires no supervision or parallel corpora, only two pretrained language models: forward and backward. Reflective Decoding operates in two intuitive steps. In the contextualization step, we use LMs to generate many left and right contexts which collectively capture the meaning of the input sentence. Then, in the reflection step we decode in the semantic neighborhood of the input, conditioning on an ensemble of generated contexts with the reverse direction LM. We reflect through the generated contexts, effectively using them as an intermediate meaning representation to generate conditional output. Empirical results demonstrate that Reflective Decoding outperforms strong unsupervised baselines on both paraphrasing and abductive text infilling, significantly narrowing the gap between unsupervised and supervised methods. Reflective Decoding introduces the concept of using generated contexts to represent meaning, opening up new possibilities for unsupervised conditional text generation. | 2withdrawn
|
Title: Plan2Vec: Unsupervised Representation Learning by Latent Plans. Abstract: Creating a useful representation of the world takes more than just rote memorization of individual data samples. This is because fundamentally, we use our internal representation to plan, to solve problems, and to navigate the world. For a representation to be amenable to planning, it is critical for it to embody some notion of optimality. A representation learning objective that explicitly considers some form of planning should generate representations which are more computationally valuable than those that memorize samples. In this paper, we introduce \textbf{Plan2Vec}, an unsupervised representation learning objective inspired by value-based reinforcement learning methods. By abstracting away low-level control with a learned local metric, we show that it is possible to learn plannable representations that inform long-range structures, entirely passively from high-dimensional sequential datasets without supervision. A latent space is learned by playing an ``Imagined Planning Game" on the graph formed by the data points, using a local metric function trained contrastively from context. We show that the global metric on this learned embedding can be used to plan with O(1) complexity by linear interpolation. This exponential speed-up is critical for planning with a learned representation on any problem containing non-trivial global topology. We demonstrate the effectiveness of Plan2Vec on simulated toy tasks from both proprioceptive and image states, as well as two real-world image datasets, showing that Plan2Vec can effectively plan using learned representations. Additional results and videos can be found at \url{https://sites.google.com/view/plan2vec}. | 0reject
|
Title: Playing the lottery with rewards and multiple languages: lottery tickets in RL and NLP. Abstract: The lottery ticket hypothesis proposes that over-parameterization of deep neural networks (DNNs) aids training by increasing the probability of a “lucky” sub-network initialization being present rather than by helping the optimization process (Frankle& Carbin, 2019). Intriguingly, this phenomenon suggests that initialization strategies for DNNs can be improved substantially, but the lottery ticket hypothesis has only previously been tested in the context of supervised learning for natural image tasks. Here, we evaluate whether “winning ticket” initializations exist in two different domains: natural language processing (NLP) and reinforcement learning (RL).For NLP, we examined both recurrent LSTM models and large-scale Transformer models (Vaswani et al., 2017). For RL, we analyzed a number of discrete-action space tasks, including both classic control and pixel control. Consistent with workin supervised image classification, we confirm that winning ticket initializations generally outperform parameter-matched random initializations, even at extreme pruning rates for both NLP and RL. Notably, we are able to find winning ticket initializations for Transformers which enable models one-third the size to achieve nearly equivalent performance. Together, these results suggest that the lottery ticket hypothesis is not restricted to supervised learning of natural images, but rather represents a broader phenomenon in DNNs. | 1accept
|
Title: Polyphonic Music Composition: An Adversarial Inverse Reinforcement Learning Approach. Abstract: Most recent approaches to automatic music harmony composition adopt deep supervised learning to train a model using a set of human composed songs as training data. However, these approaches suffer from inherent limitations from the chosen deep learning models which may lead to unpleasing harmonies. This paper explores an alternative approach to harmony composition using a combination of novel Deep Supervised Learning, DeepReinforcement Learning and Inverse Reinforcement Learning techniques. In this novel approach, our model selects the next chord in the composition(action) based on the previous notes(states), therefore allowing us to model harmony composition as a reinforcement learning problem in which we look to maximize an overall accumulated reward. However, designing an appropriate reward function is known to be a very tricky and difficult process. To overcome this problem we propose learning a reward function from a set of human-composed tracks using Adversarial Inverse Reinforcement Learning. We start by training a Bi-axial LSTM model using supervised learning and improve upon it by tuning it using Deep Q-learning. Instead of using GANs to generate a similar music composition to human compositions directly, we adopt GANs to learn the reward function of the music trajectories from human compositions. We then combine the learned reward function with a reward based on music theory rules to improve the generation of the model trained by supervised learning. The results show improvement over a pre-trained model without training with reinforcement learning with respect to a set of objective metrics and preference from subjective user evaluation. | 0reject
|
Title: Group Equivariant Generative Adversarial Networks. Abstract: Recent improvements in generative adversarial visual synthesis incorporate real and fake image transformation in a self-supervised setting, leading to increased stability and perceptual fidelity. However, these approaches typically involve image augmentations via additional regularizers in the GAN objective and thus spend valuable network capacity towards approximating transformation equivariance instead of their desired task. In this work, we explicitly incorporate inductive symmetry priors into the network architectures via group-equivariant convolutional networks. Group-convolutions have higher expressive power with fewer samples and lead to better gradient feedback between generator and discriminator. We show that group-equivariance integrates seamlessly with recent techniques for GAN training across regularizers, architectures, and loss functions. We demonstrate the utility of our methods for conditional synthesis by improving generation in the limited data regime across symmetric imaging datasets and even find benefits for natural images with preferred orientation. | 1accept
|
Title: Through the Lens of Neural Network: Analyzing Neural QA Models via Quantized Latent Representation. Abstract: In recent years, deep learning models remain black boxes, where the decision-making process is still opaque to humans. In this work, we try to explore the probabilities of understanding how machine thinks when doing question-answering tasks. In general, words are represented by continuous latent representations in the neural-based QA models. Here we train the QA models with discrete latent representations, so each word in the context is also a token in the model. In this way, we can know what a word sequence in the context looks like through the lens of the QA models. We analyze the QA models trained on QuAC (Question Answering in Context) and CoQA (A Conversational Question Answering Challenge) and organize several rules the models obey when dealing with this kind of QA task. We also find that the models maintain much of the original performance after some hidden layers are quantized. | 2withdrawn
|
Title: Wavelet Feature Maps Compression for Low Bandwidth Convolutional Neural Networks. Abstract: Quantization is one of the most effective techniques for compressing Convolutional Neural Networks (CNNs), which are known for requiring extensive computational resources. However, aggressive quantization may cause severe degradation in the prediction accuracy of such networks, especially in image-to-image tasks such as semantic segmentation and depth prediction. In this paper, we propose Wavelet Compressed Convolution (WCC)---a novel approach for activation maps compression for $1\times1$ convolutions (the workhorse of modern CNNs). WCC achieves compression ratios and computational savings that are equivalent to low bit quantization rates at a relatively minimal loss of accuracy. To this end, we use a hardware-friendly Haar-wavelet transform, known for its effectiveness in image compression, and define the convolution on the compressed activation map. WCC can be utilized with any $1\times1$ convolution in an existing network architecture. By combining WCC with light quantization, we show that we achieve compression rates equal to 2-bit and 1-bit with minimal degradation in image-to-image tasks. | 0reject
|
Title: Rectified Gradient: Layer-wise Thresholding for Sharp and Coherent Attribution Maps. Abstract: Saliency map, or the gradient of the score function with respect to the input, is the most basic means of interpreting deep neural network decisions. However, saliency maps are often visually noisy. Although several hypotheses were proposed to account for this phenomenon, there is no work that provides a rigorous analysis of noisy saliency maps. This may be a problem as numerous advanced attribution methods were proposed under the assumption that the existing hypotheses are true. In this paper, we identify the cause of noisy saliency maps. Then, we propose Rectified Gradient, a simple method that significantly improves saliency maps by alleviating that cause. Experiments showed effectiveness of our method and its superiority to other attribution methods. Codes and examples for the experiments will be released in public. | 0reject
|
Title: Multiagent Reinforcement Learning in Games with an Iterated Dominance Solution. Abstract: Multiagent reinforcement learning (MARL) attempts to optimize policies of intelligent agents interacting in the same environment. However, it may fail to converge to a Nash equilibrium in some games. We study independent MARL under the more demanding solution concept of iterated elimination of strictly dominated strategies. In dominance solvable games, if players iteratively eliminate strictly dominated strategies until no further strategies can be eliminated, we obtain a single strategy profile. We show that convergence to the iterated dominance solution is guaranteed for several reinforcement learning algorithms (for multiple independent learners). We illustrate an application of our results by studying mechanism design for principal-agent problems, where a principal wishes to incentivize agents to exert costly effort in a joint project when it can only observe whether the project succeeded, but not whether agents actually exerted effort. We show that MARL converges to the desired outcome if the rewards are designed so that exerting effort is the iterated dominance solution, but fails if it is merely a Nash equilibrium. | 0reject
|
Title: Domain Invariant Adversarial Learning. Abstract: The phenomenon of adversarial examples illustrates one of the most basic vulnerabilities of deep neural networks. Among the variety of techniques introduced to surmount this inherent weakness, adversarial training has emerged as the most effective strategy to achieve robustness. Typically, this is achieved by balancing robust and natural objectives. In this work, we aim to further reduce the trade-off between robust and standard accuracy by enforcing a domain-invariant feature representation. We present a new adversarial training method, Domain Invariant Adversarial Learning (DIAL), which learns a feature representation which is both robust and domain invariant. DIAL uses a variant of Domain Adversarial Neural Network (DANN) on the natural domain and its corresponding adversarial domain. In a case where the source domain consists of natural examples and the target domain is the adversarially perturbed examples, our method learns a feature representation constrained not to discriminate between the natural and adversarial examples, and can therefore achieve a more robust representation. Our experiments indicate that our method improves both robustness and standard accuracy, when compared to other state-of-the-art adversarial training methods. | 0reject
|
Title: Towards Understanding the Regularization of Adversarial Robustness on Neural Networks. Abstract: The problem of adversarial examples has shown that modern Neural Network (NN) models could be rather fragile. Among the most promising techniques to solve the problem, one is to require the model to be {\it $\epsilon$-adversarially robust} (AR); that is, to require the model not to change predicted labels when any given input examples are perturbed within a certain range. However, it is widely observed that such methods would lead to standard performance degradation, i.e., the degradation on natural examples. In this work, we study the degradation through the regularization perspective. We identify quantities from generalization analysis of NNs; with the identified quantities we empirically find that AR is achieved by regularizing/biasing NNs towards less confident solutions by making the changes in the feature space (induced by changes in the instance space) of most layers smoother uniformly in all directions; so to a certain extent, it prevents sudden change in prediction w.r.t. perturbations. However, the end result of such smoothing concentrates samples around decision boundaries, resulting in less confident solutions, and leads to worse standard performance. Our studies suggest that one might consider ways that build AR into NNs in a gentler way to avoid the problematic regularization.
| 0reject
|
Title: Perception-Driven Curiosity with Bayesian Surprise. Abstract: Intrinsic rewards in reinforcement learning provide a powerful algorithmic capability for agents to learn how to interact with their environment in a task-generic way. However, increased incentives for motivation can come at the cost of increased fragility to stochasticity. We introduce a method for computing an intrinsic reward for curiosity using metrics derived from sampling a latent variable model used to estimate dynamics. Ultimately, an estimate of the conditional probability of observed states is used as our intrinsic reward for curiosity. In our experiments, a video game agent uses our model to autonomously learn how to play Atari games using our curiosity reward in combination with extrinsic rewards from the game to achieve improved performance on games with sparse extrinsic rewards. When stochasticity is introduced in the environment, our method still demonstrates improved performance over the baseline. | 2withdrawn
|
Title: Can recurrent neural networks warp time?. Abstract: Successful recurrent models such as long short-term memories (LSTMs) and gated recurrent units (GRUs) use \emph{ad hoc} gating mechanisms. Empirically these models have been found to improve the learning of medium to long term temporal dependencies and to help with vanishing gradient issues.
We prove that learnable gates in a recurrent model formally provide \emph{quasi-invariance to general time transformations} in the input data. We recover part of the LSTM architecture from a simple axiomatic approach.
This result leads to a new way of initializing gate biases in LSTMs and GRUs. Experimentally, this new \emph{chrono initialization} is shown to greatly improve learning of long term dependencies, with minimal implementation effort.
| 1accept
|
Title: ConCoDE: Hard-constrained Differentiable Co-Exploration Method for Neural Architectures and Hardware Accelerators. Abstract: While DNNs achieve over-human performances in a number of areas, it is often accompanied by the skyrocketing computational costs.
Co-exploration of an optimal neural architecture and its hardware accelerator is an approach of rising interest which addresses the computational cost problem, especially in low-profile systems (e.g., embedded, mobile).
The difficulty of having to search the large co-exploration space is often addressed by adopting the idea of differentiable neural architecture search.
Despite the superior search efficiency of the differentiable co-exploration, it faces a critical challenge of not being able to systematically satisfy hard constraints, such as frame rate or power budget.
To handle the hard constraint problem of differentiable co-exploration, we propose ConCoDE,
which searches for hard-constrained solutions without compromising the global design objectives.
By manipulating the gradients in the interest of the given hard constraint, high-quality solutions satisfying the constraint can be obtained.
Experimental results show that ConCoDE is able to meet the constraints even in tight conditions.
We also show that the solutions searched by ConCoDE exhibit high quality compared to those searched without any constraint. | 2withdrawn
|
Title: ADef: an Iterative Algorithm to Construct Adversarial Deformations. Abstract: While deep neural networks have proven to be a powerful tool for many recognition and classification tasks, their stability properties are still not well understood. In the past, image classifiers have been shown to be vulnerable to so-called adversarial attacks, which are created by additively perturbing the correctly classified image. In this paper, we propose the ADef algorithm to construct a different kind of adversarial attack created by iteratively applying small deformations to the image, found through a gradient descent step. We demonstrate our results on MNIST with convolutional neural networks and on ImageNet with Inception-v3 and ResNet-101. | 1accept
|
Title: Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML. Abstract: An important research direction in machine learning has centered around developing meta-learning algorithms to tackle few-shot learning. An especially successful algorithm has been Model Agnostic Meta-Learning (MAML), a method that consists of two optimization loops, with the outer loop finding a meta-initialization, from which the inner loop can efficiently learn new tasks. Despite MAML's popularity, a fundamental open question remains -- is the effectiveness of MAML due to the meta-initialization being primed for rapid learning (large, efficient changes in the representations) or due to feature reuse, with the meta initialization already containing high quality features? We investigate this question, via ablation studies and analysis of the latent representations, finding that feature reuse is the dominant factor. This leads to the ANIL (Almost No Inner Loop) algorithm, a simplification of MAML where we remove the inner loop for all but the (task-specific) head of the underlying neural network. ANIL matches MAML's performance on benchmark few-shot image classification and RL and offers computational improvements over MAML. We further study the precise contributions of the head and body of the network, showing that performance on the test tasks is entirely determined by the quality of the learned features, and we can remove even the head of the network (the NIL algorithm). We conclude with a discussion of the rapid learning vs feature reuse question for meta-learning algorithms more broadly. | 1accept
|
Title: Variational Smoothing in Recurrent Neural Network Language Models. Abstract: We present a new theoretical perspective of data noising in recurrent neural network language models (Xie et al., 2017). We show that each variant of data noising is an instance of Bayesian recurrent neural networks with a particular variational distribution (i.e., a mixture of Gaussians whose weights depend on statistics derived from the corpus such as the unigram distribution). We use this insight to propose a more principled method to apply at prediction time and propose natural extensions to data noising under the variational framework. In particular, we propose variational smoothing with tied input and output embedding matrices and an element-wise variational smoothing method. We empirically verify our analysis on two benchmark language modeling datasets and demonstrate performance improvements over existing data noising methods. | 1accept
|
Title: Rotation-invariant clustering of neuronal responses in primary visual cortex. Abstract: Similar to a convolutional neural network (CNN), the mammalian retina encodes visual information into several dozen nonlinear feature maps, each formed by one ganglion cell type that tiles the visual space in an approximately shift-equivariant manner. Whether such organization into distinct cell types is maintained at the level of cortical image processing is an open question. Predictive models building upon convolutional features have been shown to provide state-of-the-art performance, and have recently been extended to include rotation equivariance in order to account for the orientation selectivity of V1 neurons. However, generally no direct correspondence between CNN feature maps and groups of individual neurons emerges in these models, thus rendering it an open question whether V1 neurons form distinct functional clusters. Here we build upon the rotation-equivariant representation of a CNN-based V1 model and propose a methodology for clustering the representations of neurons in this model to find functional cell types independent of preferred orientations of the neurons. We apply this method to a dataset of 6000 neurons and visualize the preferred stimuli of the resulting clusters. Our results highlight the range of non-linear computations in mouse V1. | 1accept
|
Title: ``"Best-of-Many-Samples" Distribution Matching. Abstract: Generative Adversarial Networks (GANs) can achieve state-of-the-art sample quality in generative modelling tasks but suffer from the mode collapse problem. Variational Autoencoders (VAE) on the other hand explicitly maximize a reconstruction-based data log-likelihood forcing it to cover all modes, but suffer from poorer sample quality. Recent works have proposed hybrid VAE-GAN frameworks which integrate a GAN-based synthetic likelihood to the VAE objective to address both the mode collapse and sample quality issues, with limited success. This is because the VAE objective forces a trade-off between the data log-likelihood and divergence to the latent prior. The synthetic likelihood ratio term also shows instability during training. We propose a novel objective with a ``"Best-of-Many-Samples" reconstruction cost and a stable direct estimate of the synthetic likelihood. This enables our hybrid VAE-GAN framework to achieve high data log-likelihood and low divergence to the latent prior at the same time and shows significant improvement over both hybrid VAE-GANS and plain GANs in mode coverage and quality. | 0reject
|
Title: SCELMo: Source Code Embeddings from Language Models. Abstract: Continuous embeddings of tokens in computer programs have been used to support a variety of software development tools, including readability, code search, and program repair.
Contextual embeddings are common in natural language processing but have not been previously applied in software engineering.
We introduce a new set of deep contextualized word representations for computer programs based on language models.
We train a set of embeddings using the ELMo (embeddings from language models) framework of Peters et al (2018).
We investigate whether these embeddings are effective when fine-tuned for the downstream task of bug detection.
We show that even a low-dimensional embedding trained on a relatively small corpus of programs can improve a state-of-the-art machine learning system for bug detection. | 0reject
|
Title: CDT: Cascading Decision Trees for Explainable Reinforcement Learning. Abstract: Deep Reinforcement Learning (DRL) has recently achieved significant advances in various domains. However, explaining the policy of RL agents still remains an open problem due to several factors, one being the complexity of explaining neural networks decisions. Recently, a group of works have used decision-tree-based models to learn explainable policies. Soft decision trees (SDTs) and discretized differentiable decision trees (DDTs) have been demonstrated to achieve both good performance and share the benefit of having explainable policies. In this work, we further improve the results for tree-based explainable RL in both performance and explainability. Our proposal, Cascading Decision Trees (CDTs) apply representation learning on the decision path to allow richer expressivity. Empirical results show that in both situations, where CDTs are used as policy function approximators or as imitation learners to explain black-box policies, CDTs can achieve better performances with more succinct and explainable models than SDTs. As a second contribution our study reveals limitations of explaining black-box policies via imitation learning with tree-based explainable models, due to its inherent instability. | 0reject
|
Title: Mean-field Behaviour of Neural Tangent Kernel for Deep Neural Networks. Abstract: Recent work by Jacot et al. (2018) has showed that training a neural network of any kind with gradient descent in parameter space is equivalent to kernel gradient descent in function space with respect to the Neural Tangent Kernel (NTK). Lee et al. (2019) built on this result to show that the output of a neural network trained using full batch gradient descent can be approximated by a linear model for wide networks. In parallel, a recent line of studies ( Schoenhols et al. (2017), Hayou et al. (2019)) suggested that a special initialization known as the Edge of Chaos leads to good performance. In this paper, we bridge the gap between this two concepts and show the impact of the initialization and the activation function on the NTK as the network depth becomes large. We provide experiments illustrating our theoretical results. | 0reject
|
Title: Unifying Part Detection And Association For Multi-person Pose Estimation. Abstract: Current bottom-up approaches for 2D multi-person pose estimation (MPPE) detect joints collectively without distinguishing between individuals. Associating the joints into individual poses is done independently of the learning algorithm, therefore requires formulating a separate problem in a post-processing step, which relies on relaxations or sophisticated heuristics. We propose a differentiable learning-based model that performs part detection and association jointly, thereby eliminating the need for further post-processing. The approach introduces a recurrent neural network (RNN), which takes dense low-level features as input and predicts the heatmaps of a single person's joints in each iteration, then refines them in a feedback loop. In addition, the network learns a stopping criterion in order to halt once it has identified all individuals in an image, allowing it to output any number of poses. Furthermore, we introduce an efficient implementation that allows training on memory-constrained machines. The approach is evaluated on the challenging COCO and OCHuman datasets and substantially outperforms the baseline. On OCHuman, which contains severe occlusions, we achieve state-of-the-art results even compared to top-down approaches. Our results demonstrate the advantage of a learning-based detection and association framework, and the advantage of bottom-up approaches over top-down approaches in complex scenarios. | 2withdrawn
|
Title: Dataset Inference: Ownership Resolution in Machine Learning. Abstract: With increasingly more data and computation involved in their training, machine learning models constitute valuable intellectual property. This has spurred interest in model stealing, which is made more practical by advances in learning with partial, little, or no supervision. Existing defenses focus on inserting unique watermarks in a model's decision surface, but this is insufficient: the watermarks are not sampled from the training distribution and thus are not always preserved during model stealing. In this paper, we make the key observation that knowledge contained in the stolen model's training set is what is common to all stolen copies. The adversary's goal, irrespective of the attack employed, is always to extract this knowledge or its by-products. This gives the original model's owner a strong advantage over the adversary: model owners have access to the original training data. We thus introduce $\textit{dataset inference}$, the process of identifying whether a suspected model copy has private knowledge from the original model's dataset, as a defense against model stealing. We develop an approach for dataset inference that combines statistical testing with the ability to estimate the distance of multiple data points to the decision boundary. Our experiments on CIFAR10, SVHN, CIFAR100 and ImageNet show that model owners can claim with confidence greater than 99% that their model (or dataset as a matter of fact) was stolen, despite only exposing 50 of the stolen model's training points. Dataset inference defends against state-of-the-art attacks even when the adversary is adaptive. Unlike prior work, it does not require retraining or overfitting the defended model. | 1accept
|
Title: ERNIE-SPARSE: Robust Efficient Transformer Through Hierarchically Unifying Isolated Information. Abstract: Sparse Transformer has recently attracted a lot of attention since the ability for reducing the quadratic dependency on the sequence length. In this paper, we argue that two factors could affect the robustness and causing performance degradation of the Sparse Transformer. The first factor is information bottleneck sensitivity, which is caused by the key feature of Sparse Transformer — only a small number of global tokens can attend to all other tokens. The second factor is sparse pattern sensitivity, which is caused by different token connections in different sparse patterns. To address these issues, we propose a well-designed model, named ERNIE-SPARSE. It consists of two distinctive parts: (i) a Hierarchical Sparse Transformer (HST) mechanism, which introduces special tokens to sequentially model local and global information. This method is not affected by bottleneck size and improves model robustness and performance. (ii) Sparse-Attention-Oriented Regularization (SAOR) method, the first robust training method designed for Sparse Transformer, which increases model robustness by forcing the output distributions of transformers with different sparse patterns to be consistent with each other. To evaluate the effectiveness of ERNIE-SPARSE, we perform extensive evaluations. Firstly, we perform experiments on a multi-modal long sequence modeling task benchmark, Long Range Arena (LRA). Experimental results demonstrate that ERNIE-SPARSE significantly outperforms a variety of strong baseline methods including the dense attention and other efficient sparse attention methods and achieves improvements by 2.7% (55.01% vs. 57.78%). Secondly, to further show the effectiveness of our method, we pretrain ERNIE-SPARSE and verified it on 3 text classification and 2 QA downstream tasks, achieve improvements on classification benchmark by 0.83% (91.63% vs. 92.46%), on QA benchmark by 3.27% (74.7% vs. 71.43%). Experimental results continue to demonstrate its superior performance.
| 2withdrawn
|
Title: IA-MARL: Imputation Assisted Multi-Agent Reinforcement Learning for Missing Training Data. Abstract: Recently, multi-agent reinforcement learning (MARL) adopts the centralized training with decentralized execution (CTDE) framework that trains agents using the data from all agents at a centralized server while each agent takes an action from its observation. In the real world, however, the training data from some agents can be unavailable at the centralized server due to practical reasons including communication failures and security attacks (e.g., data modification), which can slow down training and harm performance. Therefore, we consider the missing training data problem in MARL, and then propose the imputation assisted multiagent reinforcement learning (IA-MARL). IA-MARL consists of two steps: 1) the imputation of missing training data, which uses generative adversarial imputation networks (GAIN), and 2) the mask-based update of the networks, which trains each agent using the training data of corresponding agent, not missed over consecutive times. In the experimental results, we explore the effects of the data missing probability, the number of agents, and the number of pre-training episodes for GAIN on the performance of IA-MARL. We show IA-MARL outperforms a decentralized approach and even can achieve the performance of MARL without missing training data when sufficient imputation accuracy is supported. Our ablation study also shows that both the mask-based update and the imputation accuracy play important roles in achieving the high performance in IA-MARL. | 0reject
|
Title: Transformers Can Do Bayesian Inference. Abstract: Currently, it is hard to reap the benefits of deep learning for Bayesian methods, which allow the explicit specification of prior knowledge and accurately capture model uncertainty. We present Prior-Data Fitted Networks (PFNs). PFNs leverage large-scale machine learning techniques to approximate a large set of posteriors. The only requirement for PFNs to work is the ability to sample from a prior distribution over supervised learning tasks (or functions). Our method restates the objective of posterior approximation as a supervised classification problem with a set-valued input: it repeatedly draws a task (or function) from the prior, draws a set of data points and their labels from it, masks one of the labels and learns to make probabilistic predictions for it based on the set-valued input of the rest of the data points. Presented with a set of samples from a new supervised learning task as input, PFNs make probabilistic predictions for arbitrary other data points in a single forward propagation, having learned to approximate Bayesian inference. We demonstrate that PFNs can near-perfectly mimic Gaussian processes and also enable efficient Bayesian inference for intractable problems, with over 200-fold speedups in multiple setups compared to current methods. We obtain strong results in very diverse areas such as Gaussian process regression, Bayesian neural networks, classification for small tabular data sets, and few-shot image classification, demonstrating the generality of PFNs. Code and trained PFNs are released at https://github.com/automl/TransformersCanDoBayesianInference. | 1accept
|
Title: Embedding Multimodal Relational Data. Abstract: Representing entities and relations in an embedding space is a well-studied approach for machine learning on relational data. Existing approaches however primarily focus on simple link structure between a finite set of entities, ignoring the variety of data types that are often used in relational databases, such as text, images, and numerical values. In our approach, we propose a multimodal embedding using different neural encoders for this variety of data, and combine with existing models to learn embeddings of the entities. We extend existing datasets to create two novel benchmarks, YAGO-10-plus and MovieLens-100k-plus, that contain additional relations such as textual descriptions and images of the original entities. We demonstrate that our model utilizes the additional information effectively to provide further gains in accuracy. Moreover, we test our learned multimodal embeddings by using them to predict missing multimodal attributes. | 2withdrawn
|
Title: Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution. Abstract: As deep reinforcement learning (RL) is applied to more tasks, there is a need to visualize and understand the behavior of learned agents. Saliency maps explain agent behavior by highlighting the features of the input state that are most relevant for the agent in taking an action. Existing perturbation-based approaches to compute saliency often highlight regions of the input that are not relevant to the action taken by the agent. Our proposed approach, SARFA (Specific and Relevant Feature Attribution), generates more focused saliency maps by balancing two aspects (specificity and relevance) that capture different desiderata of saliency. The first captures the impact of perturbation on the relative expected reward of the action to be explained. The second downweighs irrelevant features that alter the relative expected rewards of actions other than the action to be explained. We compare SARFA with existing approaches on agents trained to play board games (Chess and Go) and Atari games (Breakout, Pong and Space Invaders). We show through illustrative examples (Chess, Atari, Go), human studies (Chess), and automated evaluation methods (Chess) that SARFA generates saliency maps that are more interpretable for humans than existing approaches. For the code release and demo videos, see: https://nikaashpuri.github.io/sarfa-saliency/. | 1accept
|
Title: Error-based or target-based? A unifying framework for learning in recurrent spiking networks. Abstract: Learning in biological or artificial networks means changing the laws governing the network dynamics in order to better behave in a specific situation. In the field of supervised learning, two complementary approaches stand out: error-based and target-based learning. However, there exists no consensus on which is better suited for which task, and what is the most biologically plausible. Here we propose a comprehensive theoretical framework that includes these two frameworks as special cases. This novel theoretical formulation offers major insights into the differences between the two approaches. In particular, we show how target-based naturally emerges from error-based when the number of constraints on the target dynamics, and as a consequence on the internal network dynamics, is comparable to the degrees of freedom of the network. Moreover, given the experimental evidences on the relevance that spikes have in biological networks, we investigate the role of coding with specific patterns of spikes by introducing a parameter that defines the tolerance to precise spike timing during learning. Our approach naturally lends itself to Imitation Learning (and Behavioral Cloning in particular) and we apply it to solve relevant closed-loop tasks such as the button-and-food task, and the 2D Bipedal Walker. We show that a high dimensionality feedback structure is extremely important when it is necessary to solve a task that requires retaining memory for a long time (button-and-food). On the other hand, we find that coding with specific patterns of spikes enables optimal performances in a motor task (the 2D Bipedal Walker). Finally, we show that our theoretical formulation suggests protocols to deduce the structure of learning feedback in biological networks. | 0reject
|
Title: Zero-training Sentence Embedding via Orthogonal Basis. Abstract: We propose a simple and robust training-free approach for building sentence representations. Inspired by the Gram-Schmidt Process in geometric theory, we build an orthogonal basis of the subspace spanned by a word and its surrounding context in a sentence. We model the semantic meaning of a word in a sentence based on two aspects. One is its relatedness to the word vector subspace already spanned by its contextual words. The other is its novel semantic meaning which shall be introduced as a new basis vector perpendicular to this existing subspace. Following this motivation, we develop an innovative method based on orthogonal basis to combine pre-trained word embeddings into sentence representation. This approach requires zero training and zero parameters, along with efficient inference performance. We evaluate our approach on 11 downstream NLP tasks. Experimental results show that our model outperforms all existing zero-training alternatives in all the tasks and it is competitive to other approaches relying on either large amounts of labelled data or prolonged training time. | 0reject
|
Title: Conscious Inference for Object Detection. Abstract: Current Convolutional Neural Network (CNN)-based object detection models adopt strictly feedforward inference to predict the final detection results. However, the widely used one-way inference is agnostic to the global image context and the interplay between input image and task semantics. In this work, we present a general technique to improve off-the-shelf CNN-based object detection models in the inference stage without re-training, architecture modification or ground-truth requirements. We propose an iterative, bottom-up and top-down inference mechanism, which is named conscious inference, as it is inspired by prevalent models for human consciousness with top-down guidance and temporal persistence. While the downstream pass accumulates category-specific evidence over time, it subsequently affects the proposal calculation and the final detection. Feature activations are updated in line with no additional memory cost. Our approach advances the state of the art using popular detection models (Faster-RCNN, YOLOv2, YOLOv3) on 2D object detection and 6D object pose estimation. | 0reject
|
Title: Breaking the Expressive Bottlenecks of Graph Neural Networks. Abstract: Recently, the Weisfeiler-Lehman (WL) graph isomorphism test was used to measure the expressiveness of graph neural networks (GNNs), showing that the neighborhood aggregation GNNs were at most as powerful as 1-WL test in distinguishing graph structures. There were also improvements proposed in analogy to $k$-WL test ($k>1$). However, the aggregators in these GNNs are far from injective as required by the WL test, and suffer from weak distinguishing strength, making it become expressive bottlenecks. In this paper, we improve the expressiveness by exploring powerful aggregators. We reformulate aggregation with the corresponding aggregation coefficient matrix, and then systematically analyze the requirements of the aggregation coefficient matrix for building more powerful aggregators and even injective aggregators. It can also be viewed as the strategy for preserving the rank of hidden features, and implies that basic aggregators correspond to a special case of low-rank transformations. We also show the necessity of applying nonlinear units ahead of aggregation, which is different from most aggregation-based GNNs. Based on our theoretical analysis, we develop two GNN layers, ExpandingConv and CombConv. Experimental results show that our models significantly boost performance, especially for large and densely connected graphs. | 0reject
|
Title: On Self Modulation for Generative Adversarial Networks. Abstract: Training Generative Adversarial Networks (GANs) is notoriously challenging. We propose and study an architectural modification, self-modulation, which improves GAN performance across different data sets, architectures, losses, regularizers, and hyperparameter settings. Intuitively, self-modulation allows the intermediate feature maps of a generator to change as a function of the input noise vector. While reminiscent of other conditioning techniques, it requires no labeled data. In a large-scale empirical study we observe a relative decrease of 5%-35% in FID. Furthermore, all else being equal, adding this modification to the generator leads to improved performance in 124/144 (86%) of the studied settings. Self-modulation is a simple architectural change that requires no additional parameter tuning, which suggests that it can be applied readily to any GAN. | 1accept
|
Title: VISCOS Flows: Variational Schur Conditional Sampling with Normalizing Flows. Abstract: We present a method for conditional sampling for pre-trained normalizing flows when only part of an observation is available. We derive a lower bound to the conditioning variable log-probability using Schur complement properties in the spirit of Gaussian conditional sampling. Our derivation relies on partitioning flow's domain in such a way that the flow restrictions to subdomains remain bijective, which is crucial for the Schur complement application. Simulation from the variational conditional flow then amends to solving an equality constraint. Our contribution is three-fold: a) we provide detailed insights on the choice of variational distributions; b) we discuss how to partition the input space of the flow to preserve bijectivity property; c) we propose a set of methods to optimise the variational distribution. Our numerical results indicate that our sampling method can be successfully applied to invertible residual networks for inference and classification. | 0reject
|
Title: Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory. Abstract: Schrödinger Bridge (SB) is an entropy-regularized optimal transport problem that has received increasing attention in deep generative modeling for its mathematical flexibility compared to the Scored-based Generative Model (SGM). However, it remains unclear whether the optimization principle of SB relates to the modern training of deep generative models, which often rely on constructing log-likelihood objectives.This raises questions on the suitability of SB models as a principled alternative for generative applications. In this work, we present a novel computational framework for likelihood training of SB models grounded on Forward-Backward Stochastic Differential Equations Theory – a mathematical methodology appeared in stochastic optimal control that transforms the optimality condition of SB into a set of SDEs. Crucially, these SDEs can be used to construct the likelihood objectives for SB that, surprisingly, generalizes the ones for SGM as special cases. This leads to a new optimization principle that inherits the same SB optimality yet without losing applications of modern generative training techniques, and we show that the resulting training algorithm achieves comparable results on generating realistic images on MNIST, CelebA, and CIFAR10. Our code is available at https://github.com/ghliu/SB-FBSDE. | 1accept
|
Title: Enabling counterfactual survival analysis with balanced representations. Abstract: Balanced representation learning methods have been applied successfully to counterfactual
inference from observational data. However, approaches that account for
survival outcomes are relatively limited. Survival data are frequently encountered
across diverse medical applications, i.e., drug development, risk profiling, and clinical
trials, and such data are also relevant in fields like manufacturing (for equipment
monitoring). When the outcome of interest is time-to-event, special precautions
for handling censored events need to be taken, as ignoring censored outcomes may
lead to biased estimates. We propose a theoretically grounded unified framework
for counterfactual inference applicable to survival outcomes. Further, we formulate
a nonparametric hazard ratio metric for evaluating average and individualized
treatment effects. Experimental results on real-world and semi-synthetic datasets,
the latter which we introduce, demonstrate that the proposed approach significantly
outperforms competitive alternatives in both survival-outcome predictions and
treatment-effect estimation. | 0reject
|