ICLR 2021 论文列表

In this paper, we cast fair machine learning as invariant machine learning. We first formulate a version of individual fairness that enforces invariance on certain sensitive sets. We then design a transport-based regularizer that enforces this version of individual fairness and develop an algorithm to minimize the regularizer efficiently. Our theoretical results guarantee the proposed approach trains certifiably fair ML models. Finally, in the experimental studies we demonstrate improved fairness metrics in comparison to several recent fair training procedures on three ML tasks that are susceptible to algorithmic bias.

Relational regularized autoencoder (RAE) is a framework to learn the distribution of data by minimizing a reconstruction loss together with a relational regularization on the prior of latent space. A recent attempt to reduce the inner discrepancy between the prior and aggregated posterior distributions is to incorporate sliced fused Gromov-Wasserstein (SFG) between these distributions. That approach has a weakness since it treats every slicing direction similarly, meanwhile several directions are not useful for the discriminative task. To improve the discrepancy and consequently the relational regularization, we propose a new relational discrepancy, named spherical sliced fused Gromov Wasserstein (SSFG), that can find an important area of projections characterized by a von Mises-Fisher distribution. Then, we introduce two variants of SSFG to improve its performance. The first variant, named mixture spherical sliced fused Gromov Wasserstein (MSSFG), replaces the vMF distribution by a mixture of von Mises-Fisher distributions to capture multiple important areas of directions that are far from each other. The second variant, named power spherical sliced fused Gromov Wasserstein (PSSFG), replaces the vMF distribution by a power spherical distribution to improve the sampling time of the vMF distribution in high dimension settings. We then apply the new discrepancies to the RAE framework to achieve its new variants. Finally, we conduct extensive experiments to show that the new autoencoders have favorable performance in learning latent manifold structure, image generation, and reconstruction.

Shape and texture are two prominent and complementary cues for recognizing objects. Nonetheless, Convolutional Neural Networks are often biased towards either texture or shape, depending on the training dataset. Our ablation shows that such bias degenerates model performance. Motivated by this observation, we develop a simple algorithm for shape-texture debiased learning. To prevent models from exclusively attending on a single cue in representation learning, we augment training data with images with conflicting shape and texture information (eg, an image of chimpanzee shape but with lemon texture) and, most importantly, provide the corresponding supervisions from shape and texture simultaneously. Experiments show that our method successfully improves model performance on several image recognition benchmarks and adversarial robustness. For example, by training on ImageNet, it helps ResNet-152 achieve substantial improvements on ImageNet (+1.2%), ImageNet-A (+5.2%), ImageNet-C (+8.3%) and Stylized-ImageNet (+11.1%), and on defending against FGSM adversarial attacker on ImageNet (+14.4%). Our method also claims to be compatible with other advanced data augmentation strategies, eg, Mixup, and CutMix. The code is available here: https://github.com/LiYingwei/ShapeTextureDebiasedTraining.

Dimensionality reduction methods for count data are critical to a wide range of applications in medical informatics and other fields where model interpretability is paramount. For such data, hierarchical Poisson matrix factorization (HPF) and other sparse probabilistic non-negative matrix factorization (NMF) methods are considered to be interpretable generative models. They consist of sparse transformations for decoding their learned representations into predictions. However, sparsity in representation decoding does not necessarily imply sparsity in the encoding of representations from the original data features. HPF is often incorrectly interpreted in the literature as if it possesses encoder sparsity. The distinction between decoder sparsity and encoder sparsity is subtle but important. Due to the lack of encoder sparsity, HPF does not possess the column-clustering property of classical NMF -- the factor loading matrix does not sufficiently define how each factor is formed from the original features. We address this deficiency by self-consistently enforcing encoder sparsity, using a generalized additive model (GAM), thereby allowing one to relate each representation coordinate to a subset of the original data features. In doing so, the method also gains the ability to perform feature selection. We demonstrate our method on simulated data and give an example of how encoder sparsity is of practical use in a concrete application of representing inpatient comorbidities in Medicare patients.

Training classifiers under fairness constraints such as group fairness, regularizes the disparities of predictions between the groups. Nevertheless, even though the constraints are satisfied during training, they might not generalize at evaluation time. To improve the generalizability of fair classifiers, we propose fair mixup, a new data augmentation strategy for imposing the fairness constraint. In particular, we show that fairness can be achieved by regularizing the models on paths of interpolated samples between the groups. We use mixup, a powerful data augmentation strategy to generate these interpolates. We analyze fair mixup and empirically show that it ensures a better generalization for both accuracy and fairness measurement in tabular, vision, and language benchmarks.

Object goal navigation aims to steer an agent towards a target object based on observations of the agent. It is of pivotal importance to design effective visual representations of the observed scene in determining navigation actions. In this paper, we introduce a Visual Transformer Network (VTNet) for learning informative visual representation in navigation. VTNet is a highly effective structure that embodies two key properties for visual representations: First, the relationships among all the object instances in a scene are exploited; Second, the spatial locations of objects and image regions are emphasized so that directional navigation signals can be learned. Furthermore, we also develop a pre-training scheme to associate the visual representations with navigation signals, and thus facilitate navigation policy learning. In a nutshell, VTNet embeds object and region features with their location cues as spatial-aware descriptors and then incorporates all the encoded descriptors through attention operations to achieve informative representation for navigation. Given such visual representations, agents are able to explore the correlations between visual observations and navigation actions. For example, an agent would prioritize ``turning right'' over ``turning left'' when the visual representation emphasizes on the right side of activation map. Experiments in the artificial environment AI2-Thor demonstrate that VTNet significantly outperforms state-of-the-art methods in unseen testing environments.

Exploratory analysis of time series data can yield a better understanding of complex dynamical systems. Granger causality is a practical framework for analysing interactions in sequential data, applied in a wide range of domains. In this paper, we propose a novel framework for inferring multivariate Granger causality under nonlinear dynamics based on an extension of self-explaining neural networks. This framework is more interpretable than other neural-network-based techniques for inferring Granger causality, since in addition to relational inference, it also allows detecting signs of Granger-causal effects and inspecting their variability over time. In comprehensive experiments on simulated data, we show that our framework performs on par with several powerful baseline methods at inferring Granger causality and that it achieves better performance at inferring interaction signs. The results suggest that our framework is a viable and more interpretable alternative to sparse-input neural networks for inferring Granger causality.

Natural data are often long-tail distributed over semantic classes. Existing recognition methods tackle this imbalanced classification by placing more emphasis on the tail data, through class re-balancing/re-weighting or ensembling over different data groups, resulting in increased tail accuracies but reduced head accuracies. We take a dynamic view of the training data and provide a principled model bias and variance analysis as the training data fluctuates: Existing long-tail classifiers invariably increase the model variance and the head-tail model bias gap remains large, due to more and larger confusion with hard negatives for the tail. We propose a new long-tailed classifier called RoutIng Diverse Experts (RIDE). It reduces the model variance with multiple experts, reduces the model bias with a distribution-aware diversity loss, reduces the computational cost with a dynamic expert routing module. RIDE outperforms the state-of-the-art by 5% to 7% on CIFAR100-LT, ImageNet-LT and iNaturalist 2018 benchmarks. It is also a universal framework that is applicable to various backbone networks, long-tailed algorithms and training mechanisms for consistent performance gains. Our code is available at: https://github.com/frank-xwang/RIDE-LongTailRecognition.

Most recent few-shot learning (FSL) approaches are based on episodic training whereby each episode samples few training instances (shots) per class to imitate the test condition. However, this strict adhering to test condition has a negative side effect, that is, the trained model is susceptible to the poor sampling of few shots. In this work, for the first time, this problem is addressed by exploiting inter-episode relationships. Specifically, a novel meta-learning via modeling episode-level relationships (MELR) framework is proposed. By sampling two episodes containing the same set of classes for meta-training, MELR is designed to ensure that the meta-learned model is robust against the presence of poorly-sampled shots in the meta-test stage. This is achieved through two key components: (1) a Cross-Episode Attention Module (CEAM) to improve the ability of alleviating the effects of poorly-sampled shots, and (2) a Cross-Episode Consistency Regularization (CECR) to enforce that the two classifiers learned from the two episodes are consistent even when there are unrepresentative instances. Extensive experiments for non-transductive standard FSL on two benchmarks show that our MELR achieves 1.0%-5.0% improvements over the baseline (i.e., ProtoNet) used for FSL in our model and outperforms the latest competitors under the same settings.

The advances in neural networks have driven many companies to provide prediction services to users in a wide range of applications. However, current prediction systems raise privacy concerns regarding the user's private data. A cryptographic neural network inference service is an efficient way to allow two parties to execute neural network inference without revealing either party’s data or model. Nevertheless, existing cryptographic neural network inference services suffer from huge running latency; in particular, the latency of communication-expensive cryptographic activation function is 3 orders of magnitude higher than plaintext-domain activation function. And activations are the necessary components of the modern neural networks. Therefore, slow cryptographic activation has become the primary obstacle of efficient cryptographic inference. In this paper, we propose a new technique, called SAFENet, to enable a Secure, Accurate and Fast nEural Network inference service. To speedup secure inference and guarantee inference accuracy, SAFENet includes channel-wise activation approximation with multiple-degree options. This is implemented by keeping the most useful activation channels and replacing the remaining, less useful, channels with various-degree polynomials. SAFENet also supports mixed-precision activation approximation by automatically assigning different replacement ratios to various layer; further increasing the approximation ratio and reducing inference latency. Our experimental results show SAFENet obtains the state-of-the-art inference latency and performance, reducing latency by $38\% \sim 61\%$ or improving accuracy by $1.8\% \sim 4\%$ over prior techniques on various encrypted datasets.

This article provides theoretical insights into the inner workings of multi-task and transfer learning methods, by studying the tractable least-square support vector machine multi-task learning (LS-SVM MTL) method, in the limit of large ($p$) and numerous ($n$) data. By a random matrix analysis applied to a Gaussian mixture data model, the performance of MTL LS-SVM is shown to converge, as $n,p\to\infty$, to a deterministic limit involving simple (small-dimensional) statistics of the data. We prove (i) that the standard MTL LS-SVM algorithm is in general strongly biased and may dramatically fail (to the point that individual single-task LS-SVMs may outperform the MTL approach, even for quite resembling tasks): our analysis provides a simple method to correct these biases, and that we reveal (ii) the sufficient statistics at play in the method, which can be efficiently estimated, even for quite small datasets. The latter result is exploited to automatically optimize the hyperparameters without resorting to any cross-validation procedure. Experiments on popular datasets demonstrate that our improved MTL LS-SVM method is computationally-efficient and outperforms sometimes much more elaborate state-of-the-art multi-task and transfer learning techniques.

Neural Architecture Search (NAS) has been explosively studied to automate the discovery of top-performer neural networks. Current works require heavy training of supernet or intensive architecture evaluations, thus suffering from heavy resource consumption and often incurring search bias due to truncated training or approximations. Can we select the best neural architectures without involving any training and eliminate a drastic portion of the search cost? We provide an affirmative answer, by proposing a novel framework called \textit{training-free neural architecture search} ($\textbf{TE-NAS}$). TE-NAS ranks architectures by analyzing the spectrum of the neural tangent kernel (NTK), and the number of linear regions in the input space. Both are motivated by recent theory advances in deep networks, and can be computed without any training. We show that: (1) these two measurements imply the $\textit{trainability}$ and $\textit{expressivity}$ of a neural network; and (2) they strongly correlate with the network's actual test accuracy. Further on, we design a pruning-based NAS mechanism to achieve a more flexible and superior trade-off between the trainability and expressivity during the search. In NAS-Bench-201 and DARTS search spaces, TE-NAS completes high-quality search but only costs $\textbf{0.5}$ and $\textbf{4}$ GPU hours with one 1080Ti on CIFAR-10 and ImageNet, respectively. We hope our work to inspire more attempts in bridging between the theoretic findings of deep networks and practical impacts in real NAS applications.

Network pruning is an effective method to reduce the computational expense of over-parameterized neural networks for deployment on low-resource systems. Recent state-of-the-art techniques for retraining pruned networks such as weight rewinding and learning rate rewinding have been shown to outperform the traditional fine-tuning technique in recovering the lost accuracy (Renda et al., 2020), but so far it is unclear what accounts for such performance. In this work, we conduct extensive experiments to verify and analyze the uncanny effectiveness of learning rate rewinding. We find that the reason behind the success of learning rate rewinding is the usage of a large learning rate. Similar phenomenon can be observed in other learning rate schedules that involve large learning rates, e.g., the 1-cycle learning rate schedule (Smith et al., 2019). By leveraging the right learning rate schedule in retraining, we demonstrate a counter-intuitive phenomenon in that randomly pruned networks could even achieve better performance than methodically pruned networks (fine-tuned with the conventional approach). Our results emphasize the cruciality of the learning rate schedule in pruned network retraining - a detail often overlooked by practitioners during the implementation of network pruning.

Recent self-supervised contrastive methods have been able to produce impressive transferable visual representations by learning to be invariant to different data augmentations. However, these methods implicitly assume a particular set of representational invariances (e.g., invariance to color), and can perform poorly when a downstream task violates this assumption (e.g., distinguishing red vs. yellow cars). We introduce a contrastive learning framework which does not require prior knowledge of specific, task-dependent invariances. Our model learns to capture varying and invariant factors for visual representations by constructing separate embedding spaces, each of which is invariant to all but one augmentation. We use a multi-head network with a shared backbone which captures information across each augmentation and alone outperforms all baselines on downstream tasks. We further find that the concatenation of the invariant and varying spaces performs best across all tasks we investigate, including coarse-grained, fine-grained, and few-shot downstream classification tasks, and various data corruptions.

Graph Convolutional Networks (GCNs) are leading methods for learning graph representations. However, without specially designed architectures, the performance of GCNs degrades quickly with increased depth. As the aggregated neighborhood size and neural network depth are two completely orthogonal aspects of graph representation, several methods focus on summarizing the neighborhood by aggregating K-hop neighborhoods of nodes while using shallow neural networks. However, these methods still encounter oversmoothing, and suffer from high computation and storage costs. In this paper, we use a modified Markov Diffusion Kernel to derive a variant of GCN called Simple Spectral Graph Convolution (SSGC). Our spectral analysis shows that our simple spectral graph convolution used in SSGC is a trade-off of low- and high-pass filter bands which capture the global and local contexts of each node. We provide two theoretical claims which demonstrate that we can aggregate over a sequence of increasingly larger neighborhoods compared to competitors while limiting severe oversmoothing. Our experimental evaluations show that SSGC with a linear learner is competitive in text and node classification tasks. Moreover, SSGC is comparable to other state-of-the-art methods for node clustering and community prediction tasks.

Powered by innovations in novel architecture design, noise tolerance techniques and increasing model capacity, Automatic Speech Recognition (ASR) has made giant strides in reducing word-error-rate over the past decade. ASR models are often trained with tens of thousand hours of high quality speech data to produce state-of-the-art (SOTA) results. Industry-scale ASR model training thus remains computationally heavy and time-consuming, and consequently has attracted little attention in adopting automatic techniques. On the other hand, Neural Architecture Search (NAS) has gained a lot of interest in the recent years thanks to its successes in discovering efficient architectures, often outperforming handcrafted alternatives. However, by changing the standard training process into a bi-level optimisation problem, NAS approaches often require significantly more time and computational power compared to single-model training, and at the same time increase complexity of the overall process. As a result, NAS has been predominately applied to problems which do not require as extensive training as ASR, and even then reproducibility of NAS algorithms is often problematic. Lately, a number of benchmark datasets has been introduced to address reproducibility issues by pro- viding NAS researchers with information about performance of different models obtained through exhaustive evaluation. However, these datasets focus mainly on computer vision and NLP tasks and thus suffer from limited coverage of application domains. In order to increase diversity in the existing NAS benchmarks, and at the same time provide systematic study of the effects of architectural choices for ASR, we release NAS-Bench-ASR – the first NAS benchmark for ASR models. The dataset consists of 8, 242 unique models trained on the TIMIT audio dataset for three different target epochs, and each starting from three different initializations. The dataset also includes runtime measurements of all the models on a diverse set of hardware platforms. Lastly, we show that identified good cell structures in our search space for TIMIT transfer well to a much larger LibriSpeech dataset.

We consider the question: how can you sample good negative examples for contrastive learning? We argue that, as with metric learning, learning contrastive representations benefits from hard negative samples (i.e., points that are difficult to distinguish from an anchor point). The key challenge toward using hard negatives is that contrastive methods must remain unsupervised, making it infeasible to adopt existing negative sampling strategies that use label information. In response, we develop a new class of unsupervised methods for selecting hard negative samples where the user can control the amount of hardness. A limiting case of this sampling results in a representation that tightly clusters each class, and pushes different classes as far apart as possible. The proposed method improves downstream performance across multiple modalities, requires only few additional lines of code to implement, and introduces no computational overhead.

Quantum Computing based Machine Learning mainly focuses on quantum computing hardware that is experimentally challenging to realize due to requiring quantum gates that operate at very low temperature. We demonstrate the existence of a "quantum computing toy model" that illustrates key aspects of quantum information processing while being experimentally accessible with room temperature optics. Pondering the question of the theoretical classification accuracy performance limit for MNIST (respectively "Fashion-MNIST") classifiers, subject to the constraint that a decision has to be made after detection of the very first photon that passed through an image-filter, we show that a machine learning system that is permitted to use quantum interference on the photon's state can substantially outperform any machine learning system that can not. Specifically, we prove that a "classical" MNIST (respectively "Fashion-MNIST") classifier cannot achieve an accuracy of better than $21.28\%$ (respectively $18.28\%$ for "Fashion-MNIST") if it must make a decision after seeing a single photon falling on one of the $28\times 28$ image pixels of a detector array. We further demonstrate that a classifier that is permitted to employ quantum interference by optically transforming the photon state prior to detection can achieve a classification accuracy of at least $41.27\%$ for MNIST (respectively $36.14\%$ for "Fashion-MNIST"). We show in detail how to train the corresponding quantum state transformation with TensorFlow and also explain how this example can serve as a teaching tool for the measurement process in quantum mechanics.

Deep reinforcement learning primarily focuses on learning behavior, usually overlooking the fact that an agent's function is largely determined by form. So, how should one go about finding a morphology fit for solving tasks in a given environment? Current approaches that co-adapt morphology and behavior use a specific task's reward as a signal for morphology optimization. However, this often requires expensive policy optimization and results in task-dependent morphologies that are not built to generalize. In this work, we propose a new approach, Task-Agnostic Morphology Evolution (TAME), to alleviate both of these issues. Without any task or reward specification, TAME evolves morphologies by only applying randomly sampled action primitives on a population of agents. This is accomplished using an information-theoretic objective that efficiently ranks agents by their ability to reach diverse states in the environment and the causality of their actions. Finally, we empirically demonstrate that across 2D, 3D, and manipulation environments TAME can evolve morphologies that match the multi-task performance of those learned with task supervised algorithms. Our code and videos can be found at https://sites.google.com/view/task-agnostic-evolution .