ICLR 2023 Papers

In this work we introduce a novel optimization algorithm for image recovery under learned sparse and low-rank constraints, which are parameterized with weighted extensions of the $\ell_p^p$-vector and $\mathcal{S}_p^p$ Schatten-matrix quasi-norms for $0\!<p\!\le1$, respectively. Our proposed algorithm generalizes the Iteratively Reweighted Least Squares (IRLS) method, used for signal recovery under $\ell_1$ and nuclear-norm constrained minimization. Further, we interpret our overall minimization approach as a recurrent network that we then employ to deal with inverse low-level computer vision problems. Thanks to the convergence guarantees that our IRLS strategy offers, we are able to train the derived reconstruction networks using a memory-efficient implicit back-propagation scheme, which does not pose any restrictions on their effective depth. To assess our networks' performance, we compare them against other existing reconstruction methods on several inverse problems, namely image deblurring, super-resolution, demosaicking and sparse recovery. Our reconstruction results are shown to be very competitive and in many cases outperform those of existing unrolled networks, whose number of parameters is orders of magnitude higher than that of our learned models.

This paper studies federated learning (FL)—especially cross-silo FL—with data from people who do not trust the server or other silos. In this setting, each silo (e.g. hospital) has data from different people (e.g. patients) and must maintain the privacy of each person’s data (e.g. medical record), even if the server or other silos act as adversarial eavesdroppers. This requirement motivates the study of Inter-Silo Record-Level Differential Privacy (ISRL-DP), which requires silo $i$’s communications to satisfy record/item-level differential privacy (DP). ISRL-DP ensures that the data of each person (e.g. patient) in silo $i$ (e.g. hospital $i$) cannot be leaked. ISRL-DP is different from well-studied privacy notions. Central and user-level DP assume that people trust the server/other silos. On the other end of the spectrum, local DP assumes that people do not trust anyone at all (even their own silo). Sitting between central and local DP, ISRL-DP makes the realistic assumption (in cross-silo FL) that people trust their own silo, but not the server or other silos. In this work, we provide tight (up to logarithms) upper and lower bounds for ISRL-DP FL with convex/strongly convex loss functions and homogeneous (i.i.d.) silo data. Remarkably, we show that similar bounds are attainable for smooth losses with arbitrary heterogeneous silo data distributions, via an accelerated ISRL-DP algorithm. We also provide tight upper and lower bounds for ISRL-DP federated empirical risk minimization, and use acceleration to attain the optimal bounds in fewer rounds of communication than the state-of-the-art. Finally, with a secure “shuffler” to anonymize silo messages (but without a trusted server), our algorithm attains the optimal central DP rates under more practical trust assumptions. Numerical experiments show favorable privacy-accuracy tradeoffs for our algorithm in classification and regression tasks.

Feature selection is the problem of selecting a subset of features for a machine learning model that maximizes model quality subject to a budget constraint. For neural networks, prior methods, including those based on $\ell_1$ regularization, attention, and other techniques, typically select the entire feature subset in one evaluation round, ignoring the residual value of features during selection, i.e., the marginal contribution of a feature given that other features have already been selected. We propose a feature selection algorithm called Sequential Attention that achieves state-of-the-art empirical results for neural networks. This algorithm is based on an efficient one-pass implementation of greedy forward selection and uses attention weights at each step as a proxy for feature importance. We give theoretical insights into our algorithm for linear regression by showing that an adaptation to this setting is equivalent to the classical Orthogonal Matching Pursuit (OMP) algorithm, and thus inherits all of its provable guarantees. Our theoretical and empirical analyses offer new explanations towards the effectiveness of attention and its connections to overparameterization, which may be of independent interest.

This paper focuses on computing the convex conjugate operation that arises when solving Euclidean Wasserstein-2 optimal transport problems. This conjugation, which is also referred to as the Legendre-Fenchel conjugate or c-transform,is considered difficult to compute and in practice,Wasserstein-2 methods are limited by not being able to exactly conjugate the dual potentials in continuous space. To overcome this, the computation of the conjugate can be approximated with amortized optimization, which learns a model to predict the conjugate. I show that combining amortized approximations to the conjugate with a solver for fine-tuning significantly improves the quality of transport maps learned for the Wasserstein-2 benchmark by Korotin et al. (2021a) and is able to model many 2-dimensional couplings and flows considered in the literature. All of the baselines, methods, and solvers in this paper are available at http://github.com/facebookresearch/w2ot.

This work analyzes the solution trajectory of gradient-based algorithms via a novel basis function decomposition. We show that, although solution trajectories of gradient-based algorithms may vary depending on the learning task, they behave almost monotonically when projected onto an appropriate orthonormal function basis. Such projection gives rise to a basis function decomposition of the solution trajectory. Theoretically, we use our proposed basis function decomposition to establish the convergence of gradient descent (GD) on several representative learning tasks. In particular, we improve the convergence of GD on symmetric matrix factorization and provide a completely new convergence result for the orthogonal symmetric tensor decomposition. Empirically, we illustrate the promise of our proposed framework on realistic deep neural networks (DNNs) across different architectures, gradient-based solvers, and datasets. Our key finding is that gradient-based algorithms monotonically learn the coefficients of a particular orthonormal function basis of DNNs defined as the eigenvectors of the conjugate kernel after training.

Semi-supervised learning (SSL) provides an effective means of leveraging unlabelled data to improve a model’s performance. Even though the domain has received a considerable amount of attention in the past years, most methods present the common drawback of lacking theoretical guarantees. Our starting point is to notice that the estimate of the risk that most discriminative SSL methods minimise is biased, even asymptotically. This bias impedes the use of standard statistical learning theory and can hurt empirical performance. We propose a simple way of removing the bias. Our debiasing approach is straightforward to implement and applicable to most deep SSL methods. We provide simple theoretical guarantees on the trustworthiness of these modified methods, without having to rely on the strong assumptions on the data distribution that SSL theory usually requires. In particular, we provide generalisation error bounds for the proposed methods. We evaluate debiased versions of different existing SSL methods, such as the Pseudo-label method and Fixmatch, and show that debiasing can compete with classic deep SSL techniques in various settings by providing better calibrated models. Additionally, we provide a theoretical explanation of the intuition of the popular SSL methods. An implementation of a debiased version of Fixmatch is available at https://github.com/HugoSchmutz/DeFixmatch

The recent success of Vision Transformers is shaking the long dominance of Convolutional Neural Networks (CNNs) in image recognition for a decade. Specifically, in terms of robustness on out-of-distribution samples, recent research finds that Transformers are inherently more robust than CNNs, regardless of different training setups. Moreover, it is believed that such superiority of Transformers should largely be credited to their \emph{self-attention-like architectures per se}. In this paper, we question that belief by closely examining the design of Transformers. Our findings lead to three highly effective architecture designs for boosting robustness, yet simple enough to be implemented in several lines of code, namely a) patchifying input images, b) enlarging kernel size, and c) reducing activation layers and normalization layers. Bringing these components together, we are able to build pure CNN architectures without any attention-like operations that are as robust as, or even more robust than, Transformers. We hope this work can help the community better understand the design of robust neural architectures. The code is publicly available at https://github.com/UCSC-VLAA/RobustCNN.

This paper studies a curious phenomenon that machine learning model with Transformer architectures have sparse activation maps. By activation map we refer to the intermediate output of the multi-layer perceptrons (MLPs) after a ReLU activation function, and by "sparse" we mean that on average very few entries (e.g., 3.0% for T5-Base and 6.3% for ViT-B16) are nonzero for each input to MLP. Moreover, larger Transformers with more layers and wider MLP hidden dimensions are sparser as measured by the percentage of nonzero entries. Through extensive experiments we demonstrate that the emergence of sparsity is a prevalent phenomenon that occurs for both natural language processing and vision tasks, on both training and evaluation data, for Transformers of various configurations, at layers of all depth levels. We discuss how sparsity immediately implies a way to significantly reduce the FLOP count and improve efficiency for Transformers. Moreover, we demonstrate perhaps surprisingly that enforcing an even sparser activation via Top-k thresholding with a small k brings a collection of desired properties, namely less sensitivity to noisy training data, more robustness to input corruptions, and better calibration for their prediction confidence.

Convolutional models have been widely used in multiple domains. However, most existing models only use local convolution, making the model unable to handle long-range dependencies efficiently. Attention overcomes this problem by aggregating global information based on the pair-wise attention score but also makes the computational complexity quadratic to the sequence length. Recently, Gu et al. proposed a model called S4 inspired by the state space model. S4 can be efficiently implemented as a global convolutional model whose kernel size equals the input sequence length. With Fast Fourier Transform, S4 can model much longer sequences than Transformers and achieve significant gains over SoTA on several long-range tasks. Despite its empirical success, S4 is involved. It requires sophisticated parameterization and initialization schemes that combine the wisdom from several prior works. As a result, S4 is less intuitive and hard to use for researchers with limited prior knowledge. Here we aim to demystify S4 and extract basic principles that contribute to the success of S4 as a global convolutional model. We focus on the structure of the convolution kernel and identify two critical but intuitive principles enjoyed by S4 that are sufficient to make up an effective global convolutional model: 1) The parameterization of the convolutional kernel needs to be efficient in the sense that the number of parameters should scale sub-linearly with sequence length. 2) The kernel needs to satisfy a decaying structure that the weights for convolving with closer neighbors are larger than the more distant ones. Based on the two principles, we propose a simple yet effective convolutional model called Structured Global Convolution (SGConv). SGConv exhibits strong empirical performance over several tasks: 1) With faster speed, SGConv surpasses the previous SoTA on Long Range Arena and Speech Command datasets. 2) When plugging SGConv into standard language and vision models, it shows the potential to improve both efficiency and performance.

Understanding dynamics from visual observations is a challenging problem that requires disentangling individual objects from the scene and learning their interactions. While recent object-centric models can successfully decompose a scene into objects, modeling their dynamics effectively still remains a challenge. We address this problem by introducing SlotFormer -- a Transformer-based autoregressive model operating on learned object-centric representations. Given a video clip, our approach reasons over object features to model spatio-temporal relationships and predicts accurate future object states. In this paper, we successfully apply SlotFormer to perform video prediction on datasets with complex object interactions. Moreover, the unsupervised SlotFormer's dynamics model can be used to improve the performance on supervised downstream tasks, such as Visual Question Answering (VQA), and goal-conditioned planning. Compared to past works on dynamics modeling, our method achieves significantly better long-term synthesis of object dynamics, while retaining high quality visual generation. Besides, SlotFormer enables VQA models to reason about the future without object-level labels, even outperforming counterparts that use ground-truth annotations. Finally, we show its ability to serve as a world model for model-based planning, which is competitive with methods designed specifically for such tasks.

Existing federated learning paradigms usually extensively exchange distributed models, rather than original data, at a central solver to achieve a more powerful model. However, this would incur severe communication burden between a server and multiple clients especially when data distributions are heterogeneous. As a result, current federated learning methods often require plenty of communication rounds in training. Unlike existing paradigms, we introduce an alternative perspective to significantly decrease the federate learning communication cost without leaking original data. In this work, we first present a meta knowledge representation method that extracts meta knowledge from distributed clients. The extracted meta knowledge encodes essential information that can be used to improve the current model. As the training progresses, the contributions of the same training samples to a federated model should also vary. Thus, we introduce a dynamic weight assignment mechanism that enables informative samples to contribute adaptively to the current model update. Then, informative meta knowledge from all active clients is sent to the server for model update. Training model on the combined meta knowledge that is regarded as a condense form of original data can significantly mitigate the heterogeneity issues. Moreover, to further ameliorate data heterogeneity, we also exchange meta knowledge among clients as conditional initialisation for meta knowledge extraction. Extensive experiments demonstrate the effectiveness and efficiency of our proposed method. Remarkably, our method outperforms the state-of-the-art by a large margin (from $74.07\%$ to $92.95\%$) on MNIST with a restricted communication budget (\textit{i.e.}, 10 rounds).

Learning from large amounts of unsupervised data and a small amount of supervision is an important open problem in computer vision. We propose a new semi-supervised learning method, Semantic Positives via Pseudo-Labels (SEMPPL), that combines labelled and unlabelled data to learn informative representations. Our method extends self-supervised contrastive learning—where representations are shaped by distinguishing whether two samples represent the same underlying datum (positives) or not (negatives)—with a novel approach to selecting positives. To enrich the set of positives, we leverage the few existing ground-truth labels to predict the missing ones through a k-nearest neighbors classifier by using the learned embeddings of the labelled data. We thus extend the set of positives with datapoints having the same pseudo-label and call these semantic positives. We jointly learn the representation and predict bootstrapped pseudo-labels. This creates a reinforcing cycle. Strong initial representations enable better pseudo-label predictions which then improve the selection of semantic positives and lead to even better representations. SEMPPL outperforms competing semi-supervised methods setting new state-of-the-art performance of 76% and 68.5% top-1accuracy when using a ResNet-50 and training on 10% and 1% of labels on ImageNet, respectively. Furthermore, when using selective kernels, SEMPPL significantly outperforms previous state-of-the-art achieving 72.3% and 78.3% top-1accuracy on ImageNet with 1% and 10% labels, respectively, which improves absolute +7.8% and +6.2% over previous work. SEMPPL also exhibits state-of-the-art performance over larger ResNet models as well as strong robustness, out-of-distribution and transfer performance. We release the checkpoints and the evaluation code at https://github.com/deepmind/semppl.

Training large, deep neural networks to convergence can be prohibitively expensive. As a result, often only a small selection of popular, dense models are reused across different contexts and tasks. Increasingly, sparsely activated models, which seek to decouple model size from computation costs, are becoming an attractive alternative to dense models. Although more efficient in terms of quality and computation cost, sparse models remain data-hungry and costly to train from scratch in the large scale regime. In this work, we propose sparse upcycling -- a simple way to reuse sunk training costs by initializing a sparsely activated Mixture-of-Experts model from a dense checkpoint. We show that sparsely upcycled T5 Base, Large, and XL language models and Vision Transformer Base and Large models, respectively, significantly outperform their dense counterparts on SuperGLUE and ImageNet, using only ~50% of the initial dense pretraining sunk cost. The upcycled models also outperform sparse models trained from scratch on 100% of the initial dense pretraining computation budget.

Fast and accurate adaptation of automatic speech recognition (ASR) systems using only text data in the target domain is a problem of long-standing practical relevance. Text-only adaptation was easy in traditional cascaded ASR systems with completely decoupled acoustic and language models. Recently, the RNNTransducer (RNN-T) has emerged as a default ASR model because of its high accuracy, low latency, and capability of supporting streaming input. However text-only adaptation of the RNN-T model is significantly more challenging due to its tight integration of acoustic and language models and end-to-end training. Existing recent approaches for text-only adaptation of RNN-Ts, either entail significant modification to the network or introduce high latency during decoding. We propose a new approach (TOLSTOI) that imputes speech representations internal to a baseline RNN-T, starting from text-only inputs, and performs in-situ adaptation that results in higher adaptation accuracy without any runtime overheads during decoding. Our imputation model is a function of the labeled data and trained parameters of the ASR model, and that we show, is more effective in controlling catastrophic forgetting compared to existing methods. We establish the effectiveness of TOLSTOI using three target domains and two ASR models of varying complexity. We yield up to 35% relative reduction in word error rate with text-only adaptation while forgetting the least compared to existing adaptation approaches. Our method is easy to implement and can be harnessed on existing RNN-T models without requiring ASR model training from scratch.

Recently, RL has shown its strong ability for visually complex tasks. However, it suffers from the low sample efficiency and poor generalization ability, which prevent RL from being useful in real-world scenarios. Inspired by the huge success of unsupervised pre-training methods on language and vision domains, we propose to improve the sample efficiency via a novel pre-training method for model-based RL. Instead of using pre-recorded agent trajectories that come with their own actions, we consider the setting where the pre-training data are action-free videos, which are more common and available in the real world. We introduce a two-phase training pipeline as follows: for the pre-training phase, we implicitly extract the hidden action embedding from videos and pre-train the visual representation and the environment dynamics network through a novel \Changes{forward-inverse} cycle consistency \Changes{(FICC)} objective based on vector quantization; for down-stream tasks, we finetune with small amount of task data based on the learned models. Our framework can significantly improve the sample efficiency on Atari Games with data of only one hour of game playing. We achieve 118.4\% mean human performance and 36.0\% median performance with only 50k environment steps, which is 85.6\% and 65.1\% better than the scratch EfficientZero model. We believe such pre-training approach can provide an option for solving real-world RL problems. The code is available at \url{https://github.com/YeWR/FICC.git}.

Complex Query Answering (CQA) over Knowledge Graphs (KGs) has attracted a lot of attention to potentially support many applications. Given that KGs are usually incomplete, neural models are proposed to answer the logical queries by parameterizing set operators with complex neural networks. However, such methods usually train neural set operators with a large number of entity and relation embeddings from the zero, where whether and how the embeddings or the neural set operators contribute to the performance remains not clear. In this paper, we propose a simple framework for complex query answering that decomposes the KG embeddings from neural set operators. We propose to represent the complex queries into the query graph. On top of the query graph, we propose the Logical Message Passing Neural Network (LMPNN) that connects the local one-hop inferences on atomic formulas to the global logical reasoning for complex query answering. We leverage existing effective KG embeddings to conduct one-hop inferences on atomic formulas, the results of which are regarded as the messages passed in LMPNN. The reasoning process over the overall logical formulas is turned into the forward pass of LMPNN that incrementally aggregates local information to finally predict the answers' embeddings. The complex logical inference across different types of queries will then be learned from training examples based on the LMPNN architecture. Theoretically, our query-graph represenation is more general than the prevailing operator-tree formulation, so our approach applies to a broader range of complex KG queries. Empirically, our approach yields the new state-of-the-art neural CQA model. Our research bridges the gap between complex KG query answering tasks and the long-standing achievements of knowledge graph representation learning. Our implementation can be found at https://github.com/HKUST-KnowComp/LMPNN.

AlphaZero-type algorithms may stop improving on single-player tasks in case the value network guiding the tree search is unable to approximate the outcome of an episode sufficiently well. One technique to address this problem is transforming the single-player task through self-competition. The main idea is to compute a scalar baseline from the agent’s historical performances and to reshape an episode’s reward into a binary output, indicating whether the baseline has been exceeded or not. However, this baseline only carries limited information for the agent about strategies how to improve. We leverage the idea of self-competition and directly incorporate a historical policy into the planning process instead of its scalar performance. Based on the recently introduced Gumbel AlphaZero (GAZ), we propose our algorithm GAZ ‘Play-to-Plan’ (GAZ PTP), in which the agent learns to find strong trajectories by planning against possible strategies of its past self. We show the effectiveness of our approach in two well-known combinatorial optimization problems, the Traveling Salesman Problem and the Job-Shop Scheduling Problem. With only half of the simulation budget for search, GAZ PTP consistently outperforms all selected single-player variants of GAZ.

Experience replay, which stores transitions in a replay memory for repeated use, plays an important role of improving sample efficiency in reinforcement learning. Existing techniques such as reweighted sampling, episodic learning and reverse sweep update further process the information in the replay memory to make experience replay more efficient. In this work, we further exploit the information in the replay memory by treating it as an empirical \emph{Replay Memory MDP (RM-MDP)}. By solving it with dynamic programming, we learn a conservative value estimate that \emph{only} considers transitions observed in the replay memory. Both value and policy regularizers based on this conservative estimate are developed and integrated with model-free learning algorithms. We design the metric \textit{memory density} to measure the quality of RM-MDP. Our empirical studies quantitatively find a strong correlation between performance improvement and memory density. Our method combines \emph{Conservative Estimation with Experience Replay (CEER)}, improving sample efficiency by a large margin, especially when the memory density is high. Even when the memory density is low, such a conservative estimate can still help to avoid suicidal actions and thereby improve performance.

Unsupervised text-to-speech (TTS) aims to train TTS models for a specific language without any paired speech-text training data in that language. Existing methods either use speech and corresponding pseudo text generated by an unsupervised automatic speech recognition (ASR) model as training data, or employ the back-translation technique. Though effective, they suffer from low robustness to low-quality data and heavy dependence on the lexicon of a language that is sometimes unavailable, leading to difficulty in convergence, especially in low-resource language scenarios. In this work, we introduce a bag of tricks to enable effective unsupervised TTS. Specifically, 1) we carefully design a voice conversion model to normalize the variable and noisy information in the low-quality speech data while preserving the pronunciation information; 2) we employ the non-autoregressive TTS model to overcome the robustness issue; and 3) we explore several tricks applied in back-translation, including curriculum learning, length augmentation and auxiliary supervised loss to stabilize the back-translation and improve its effectiveness. Through experiments, it has been demonstrated that our method achieves better intelligibility and audio quality than all previous methods, and that these tricks are very essential to the performance gain.