ICLR 2021 论文列表

Understanding the algorithmic bias of stochastic gradient descent (SGD) is one of the key challenges in modern machine learning and deep learning theory. Most of the existing works, however, focus on very small or even infinitesimal learning rate regime, and fail to cover practical scenarios where the learning rate is moderate and annealing. In this paper, we make an initial attempt to characterize the particular regularization effect of SGD in the moderate learning rate regime by studying its behavior for optimizing an overparameterized linear regression problem. In this case, SGD and GD are known to converge to the unique minimum-norm solution; however, with the moderate and annealing learning rate, we show that they exhibit different directional bias: SGD converges along the large eigenvalue directions of the data matrix, while GD goes after the small eigenvalue directions. Furthermore, we show that such directional bias does matter when early stopping is adopted, where the SGD output is nearly optimal but the GD output is suboptimal. Finally, our theory explains several folk arts in practice used for SGD hyperparameter tuning, such as (1) linearly scaling the initial learning rate with batch size; and (2) overrunning SGD with high learning rate even when the loss stops decreasing.

The study of deep neural networks (DNNs) in the infinite-width limit, via the so-called neural tangent kernel (NTK) approach, has provided new insights into the dynamics of learning, generalization, and the impact of initialization. One key DNN architecture remains to be kernelized, namely, the recurrent neural network (RNN). In this paper we introduce and study the Recurrent Neural Tangent Kernel (RNTK), which provides new insights into the behavior of overparametrized RNNs. A key property of the RNTK should greatly benefit practitioners is its ability to compare inputs of different length. To this end, we characterize how the RNTK weights different time steps to form its output under different initialization parameters and nonlinearity choices. A synthetic and 56 real-world data experiments demonstrate that the RNTK offers significant performance gains over other kernels, including standard NTKs, across a wide array of data sets.

Low-precision neural networks represent both weights and activations with few bits, drastically reducing the cost of multiplications. Meanwhile, these products are accumulated using high-precision (typically 32-bit) additions. Additions dominate the arithmetic complexity of inference in quantized (e.g., binary) nets, and high precision is needed to avoid overflow. To further optimize inference, we propose WrapNet, an architecture that adapts neural networks to use low-precision (8-bit) additions while achieving classification accuracy comparable to their 32-bit counterparts. We achieve resilience to low-precision accumulation by inserting a cyclic activation layer that makes results invariant to overflow. We demonstrate the efficacy of our approach using both software and hardware platforms.

The capability of incrementally learning new tasks without forgetting old ones is a challenging problem due to catastrophic forgetting. This challenge becomes greater when novel tasks contain very few labelled training samples. Currently, most methods are dedicated to class-incremental learning and rely on sufficient training data to learn additional weights for newly added classes. Those methods cannot be easily extended to incremental regression tasks and could suffer from severe overfitting when learning few-shot novel tasks. In this study, we propose a nonparametric method in deep embedded space to tackle incremental few-shot learning problems. The knowledge about the learned tasks are compressed into a small number of quantized reference vectors. The proposed method learns new tasks sequentially by adding more reference vectors to the model using few-shot samples in each novel task. For classification problems, we employ the nearest neighbor scheme to make classification on sparsely available data and incorporate intra-class variation, less forgetting regularization and calibration of reference vectors to mitigate catastrophic forgetting. In addition, the proposed learning vector quantization (LVQ) in deep embedded space can be customized as a kernel smoother to handle incremental few-shot regression tasks. Experimental results demonstrate that the proposed method outperforms other state-of-the-art methods in incremental learning.

Learning to predict the long-term future of video frames is notoriously challenging due to the inherent ambiguities in a distant future and dramatic amplification of prediction error over time. Despite the recent advances in the literature, existing approaches are limited to moderately short-term prediction (less than a few seconds), while extrapolating it to a longer future quickly leads to destruction in structure and content. In this work, we revisit the hierarchical models in video prediction. Our method generates future frames by first estimating a sequence of dense semantic structures and subsequently translating the estimated structures to pixels by video-to-video translation model. Despite the simplicity, we show that modeling structures and their dynamics in categorical structure space with stochastic sequential estimator leads to surprisingly successful long-term prediction. We evaluate our method on two challenging video prediction scenarios, \emph{car driving} and \emph{human dancing}, and demonstrate that it can generate complicated scene structures and motions over a very long time horizon (\ie~thousands frames), setting a new standard of video prediction with orders of magnitude longer prediction time than existing approaches. Video results are available at https://1konny.github.io/HVP/.

Standard Convolutional Neural Networks (CNNs) can be easily fooled by images with small quasi-imperceptible artificial perturbations. As alternatives to CNNs, the recently proposed Capsule Networks (CapsNets) are shown to be more robust to white-box attack than CNNs under popular attack protocols. Besides, the class-conditional reconstruction part of CapsNets is also used to detect adversarial examples. In this work, we investigate the adversarial robustness of CapsNets, especially how the inner workings of CapsNets change when the output capsules are attacked. The first observation is that adversarial examples misled CapsNets by manipulating the votes from primary capsules. Another observation is the high computational cost, when we directly apply multi-step attack methods designed for CNNs to attack CapsNets, due to the computationally expensive routing mechanism. Motivated by these two observations, we propose a novel vote attack where we attack votes of CapsNets directly. Our vote attack is not only effective, but also efficient by circumventing the routing process. Furthermore, we integrate our vote attack into the detection-aware attack paradigm, which can successfully bypass the class-conditional reconstruction based detection method. Extensive experiments demonstrate the superior attack performance of our vote attack on CapsNets.

Human-annotated labels are often prone to noise, and the presence of such noise will degrade the performance of the resulting deep neural network (DNN) models. Much of the literature (with several recent exceptions) of learning with noisy labels focuses on the case when the label noise is independent of features. Practically, annotations errors tend to be instance-dependent and often depend on the difficulty levels of recognizing a certain task. Applying existing results from instance-independent settings would require a significant amount of estimation of noise rates. Therefore, providing theoretically rigorous solutions for learning with instance-dependent label noise remains a challenge. In this paper, we propose CORES$^{2}$ (COnfidence REgularized Sample Sieve), which progressively sieves out corrupted examples. The implementation of CORES$^{2}$ does not require specifying noise rates and yet we are able to provide theoretical guarantees of CORES$^{2}$ in filtering out the corrupted examples. This high-quality sample sieve allows us to treat clean examples and the corrupted ones separately in training a DNN solution, and such a separation is shown to be advantageous in the instance-dependent noise setting. We demonstrate the performance of CORES$^{2}$ on CIFAR10 and CIFAR100 datasets with synthetic instance-dependent label noise and Clothing1M with real-world human noise. As of independent interests, our sample sieve provides a generic machinery for anatomizing noisy datasets and provides a flexible interface for various robust training techniques to further improve the performance. Code is available at https://github.com/UCSC-REAL/cores.

We introduce Skip-Window, a method to allow recurrent neural networks (RNNs) to trade off accuracy for computational cost during the analysis of a sequence. Similarly to existing approaches, Skip-Window extends existing RNN cells by adding a mechanism to encourage the model to process fewer inputs. Unlike existing approaches, Skip-Window is able to respect a strict computational budget, making this model more suitable for limited hardware. We evaluate this approach on two datasets: a human activity recognition task and adding task. Our results show that Skip-Window is able to exceed the accuracy of existing approaches for a lower computational cost while strictly limiting said cost.

Despite their success in massive engineering applications, deep neural networks are vulnerable to various perturbations due to their black-box nature. Recent study has shown that a deep neural network can misclassify the data even if the input data is perturbed by an imperceptible amount. In this paper, we address the robustness issue of neural networks by a novel close-loop control method from the perspective of dynamic systems. Instead of modifying the parameters in a fixed neural network architecture, a close-loop control process is added to generate control signals adaptively for the perturbed or corrupted data. We connect the robustness of neural networks with optimal control using the geometrical information of underlying data to design the control objective. The detailed analysis shows how the embedding manifolds of state trajectory affect error estimation of the proposed method. Our approach can simultaneously maintain the performance on clean data and improve the robustness against many types of data perturbations. It can also further improve the performance of robustly trained neural networks against different perturbations. To the best of our knowledge, this is the first work that improves the robustness of neural networks with close-loop control.

Neural Ordinary Differential Equations (NODEs) use a neural network to model the instantaneous rate of change in the state of a system. However, despite their apparent suitability for dynamics-governed time-series, NODEs present a few disadvantages. First, they are unable to adapt to incoming data-points, a fundamental requirement for real-time applications imposed by the natural direction of time. Second, time-series are often composed of a sparse set of measurements that could be explained by many possible underlying dynamics. NODEs do not capture this uncertainty. In contrast, Neural Processes (NPs) are a new class of stochastic processes providing uncertainty estimation and fast data-adaptation, but lack an explicit treatment of the flow of time. To address these problems, we introduce Neural ODE Processes (NDPs), a new class of stochastic processes determined by a distribution over Neural ODEs. By maintaining an adaptive data-dependent distribution over the underlying ODE, we show that our model can successfully capture the dynamics of low-dimensional systems from just a few data-points. At the same time, we demonstrate that NDPs scale up to challenging high-dimensional time-series with unknown latent dynamics such as rotating MNIST digits.

Transfer of pre-trained representations can improve sample efficiency and reduce computational requirements for new tasks. However, representations used for transfer are usually generic, and are not tailored to a particular distribution of downstream tasks. We explore the use of expert representations for transfer with a simple, yet effective, strategy. We train a diverse set of experts by exploiting existing label structures, and use cheap-to-compute performance proxies to select the relevant expert for each target task. This strategy scales the process of transferring to new tasks, since it does not revisit the pre-training data during transfer. Accordingly, it requires little extra compute per target task, and results in a speed-up of 2-3 orders of magnitude compared to competing approaches. Further, we provide an adapter-based architecture able to compress many experts into a single model. We evaluate our approach on two different data sources and demonstrate that it outperforms baselines on over 20 diverse vision tasks in both cases.

We propose a novel information bottleneck (IB) method named Drop-Bottleneck, which discretely drops features that are irrelevant to the target variable. Drop-Bottleneck not only enjoys a simple and tractable compression objective but also additionally provides a deterministic compressed representation of the input variable, which is useful for inference tasks that require consistent representation. Moreover, it can jointly learn a feature extractor and select features considering each feature dimension's relevance to the target task, which is unattainable by most neural network-based IB methods. We propose an exploration method based on Drop-Bottleneck for reinforcement learning tasks. In a multitude of noisy and reward sparse maze navigation tasks in VizDoom (Kempka et al., 2016) and DMLab (Beattie et al., 2016), our exploration method achieves state-of-the-art performance. As a new IB framework, we demonstrate that Drop-Bottleneck outperforms Variational Information Bottleneck (VIB) (Alemi et al., 2017) in multiple aspects including adversarial robustness and dimensionality reduction.

The attention mechanism has been widely used in deep neural networks as a model component. By now, it has become a critical building block in many state-of-the-art natural language models. Despite its great success established empirically, the working mechanism of attention has not been investigated at a sufficient theoretical depth to date. In this paper, we set up a simple text classification task and study the dynamics of training a simple attention-based classification model using gradient descent. In this setting, we show that, for the discriminative words that the model should attend to, a persisting identity exists relating its embedding and the inner product of its key and the query. This allows us to prove that training must converge to attending to the discriminative words when the attention output is classified by a linear classifier. Experiments are performed, which validate our theoretical analysis and provide further insights.

For many practical, high-risk applications, it is essential to quantify uncertainty in a model's predictions to avoid costly mistakes. While predictive uncertainty is widely studied for neural networks, the topic seems to be under-explored for models based on gradient boosting. However, gradient boosting often achieves state-of-the-art results on tabular data. This work examines a probabilistic ensemble-based framework for deriving uncertainty estimates in the predictions of gradient boosting classification and regression models. We conducted experiments on a range of synthetic and real datasets and investigated the applicability of ensemble approaches to gradient boosting models that are themselves ensembles of decision trees. Our analysis shows that ensembles of gradient boosting models successfully detect anomalous inputs while having limited ability to improve the predicted total uncertainty. Importantly, we also propose a concept of a virtual ensemble to get the benefits of an ensemble via only one gradient boosting model, which significantly reduces complexity.

While deep learning has been very beneficial in data-rich settings, tasks with smaller training set often resort to pre-training or multitask learning to leverage data from other tasks. In this case, careful consideration is needed to select tasks and model parameterizations such that updates from the auxiliary tasks actually help the primary task. We seek to alleviate this burden by formulating a model-agnostic framework that performs fine-grained manipulation of the auxiliary task gradients. We propose to decompose auxiliary updates into directions which help, damage or leave the primary task loss unchanged. This allows weighting the update directions differently depending on their impact on the problem of interest. We present a novel and efficient algorithm for that purpose and show its advantage in practice. Our method leverages efficient automatic differentiation procedures and randomized singular value decomposition for scalability. We show that our framework is generic and encompasses some prior work as particular cases. Our approach consistently outperforms strong and widely used baselines when leveraging out-of-distribution data for Text and Image classification tasks.

As an essential ingredient of modern deep learning, attention mechanism, especially self-attention, plays a vital role in the global correlation discovery. However, is hand-crafted attention irreplaceable when modeling the global context? Our intriguing finding is that self-attention is not better than the matrix decomposition~(MD) model developed 20 years ago regarding the performance and computational cost for encoding the long-distance dependencies. We model the global context issue as a low-rank completion problem and show that its optimization algorithms can help design global information blocks. This paper then proposes a series of Hamburgers, in which we employ the optimization algorithms for solving MDs to factorize the input representations into sub-matrices and reconstruct a low-rank embedding. Hamburgers with different MDs can perform favorably against the popular global context module self-attention when carefully coping with gradients back-propagated through MDs. Comprehensive experiments are conducted in the vision tasks where it is crucial to learn the global context, including semantic segmentation and image generation, demonstrating significant improvements over self-attention and its variants. Code is available at https://github.com/Gsunshine/Enjoy-Hamburger.

Training Generative Adversarial Networks (GAN) on high-fidelity images usually requires large-scale GPU-clusters and a vast number of training images. In this paper, we study the few-shot image synthesis task for GAN with minimum computing cost. We propose a light-weight GAN structure that gains superior quality on 1024^2 resolution. Notably, the model converges from scratch with just a few hours of training on a single RTX-2080 GPU, and has a consistent performance, even with less than 100 training samples. Two technique designs constitute our work, a skip-layer channel-wise excitation module and a self-supervised discriminator trained as a feature-encoder. With thirteen datasets covering a wide variety of image domains (The datasets and code are available at https://github.com/odegeasslbc/FastGAN-pytorch), we show our model's superior performance compared to the state-of-the-art StyleGAN2, when data and computing budget are limited.

Deep generative adversarial networks (GANs) have gained growing popularity in numerous scenarios, while usually suffer from high parameter complexities for resource-constrained real-world applications. However, the compression of GANs has less been explored. A few works show that heuristically applying compression techniques normally leads to unsatisfactory results, due to the notorious training instability of GANs. In parallel, the lottery ticket hypothesis shows prevailing success on discriminative models, in locating sparse matching subnetworks capable of training in isolation to full model performance. In this work, we for the first time study the existence of such trainable matching subnetworks in deep GANs. For a range of GANs, we certainly find matching subnetworks at $67\%$-$74\%$ sparsity. We observe that with or without pruning discriminator has a minor effect on the existence and quality of matching subnetworks, while the initialization weights used in the discriminator plays a significant role. We then show the powerful transferability of these subnetworks to unseen tasks. Furthermore, extensive experimental results demonstrate that our found subnetworks substantially outperform previous state-of-the-art GAN compression approaches in both image generation (e.g. SNGAN) and image-to-image translation GANs (e.g. CycleGAN). Codes available at https://github.com/VITA-Group/GAN-LTH.

There has been increasing interest in building deep hierarchy-aware classifiers that aim to quantify and reduce the severity of mistakes, and not just reduce the number of errors. The idea is to exploit the label hierarchy (e.g., the WordNet ontology) and consider graph distances as a proxy for mistake severity. Surprisingly, on examining mistake-severity distributions of the top-1 prediction, we find that current state-of-the-art hierarchy-aware deep classifiers do not always show practical improvement over the standard cross-entropy baseline in making better mistakes. The reason for the reduction in average mistake-severity can be attributed to the increase in low-severity mistakes, which may also explain the noticeable drop in their accuracy. To this end, we use the classical Conditional Risk Minimization (CRM) framework for hierarchy-aware classification. Given a cost matrix and a reliable estimate of likelihoods (obtained from a trained network), CRM simply amends mistakes at inference time; it needs no extra hyperparameters and requires adding just a few lines of code to the standard cross-entropy baseline. It significantly outperforms the state-of-the-art and consistently obtains large reductions in the average hierarchical distance of top-$k$ predictions across datasets, with very little loss in accuracy. CRM, because of its simplicity, can be used with any off-the-shelf trained model that provides reliable likelihood estimates.