Datasets:

kreas

/

DAML4-TLDR-GEN-DATASET

like 0

The fundamental principle of Graph Neural Networks (GNNs) is to exploit the structural information of the data by aggregating the neighboring nodes using a `graph convolution’. Therefore, understanding its influence on the network performance is crucial. Convolutions based on graph Laplacian have emerged as the dominant choice with the symmetric normalization of the adjacency matrix $A$, defined as $D^{-1/2}AD^{−1/2}$, being the most widely adopted one, where $D$ is the degree matrix. However, some empirical studies show that row normalization $D^{−1}A$ outperforms it in node classification. Despite the widespread use of GNNs, there is no rigorous theoretical study on the representation power of these convolution operators, that could explain this behavior. In this work, we analyze the influence of the graph convolutions theoretically using Graph Neural Tangent Kernel in a semi-supervised node classification setting. Under a Degree Corrected Stochastic Block Model, we prove that: (i) row normalization preserves the underlying class structure better than other graph convolutions; (ii) performance degrades with network depth due to over-smoothing, but the loss in class information is the slowest in row normalization; (iii) skip connections retain the class information even at infinite depth, thereby eliminating over-smoothing. We finally validate our theoretical findings on real datasets.

Graph NTK shows that row normalized graph convolution preserves the underlying class structure, and skip connections retain the class structure at infinite depth.

Unsupervised reinforcement learning (URL) poses a promising paradigm to learn useful behaviors in a task-agnostic environment without the guidance of extrinsic rewards to facilitate the fast adaptation of various downstream tasks. Previous works focused on the pre-training in a model-free manner while lacking the study of transition dynamics modeling that leaves a large space for the improvement of sample efficiency in downstream tasks. To this end, we propose an Efficient Unsupervised Reinforcement Learning Framework with Multi-choice Dynamics model (EUCLID), which introduces a novel model-fused paradigm to jointly pre-train the dynamics model and unsupervised exploration policy in the pre-training phase, thus better leveraging the environmental samples and improving the downstream task sampling efficiency. However, constructing a generalizable model which captures the local dynamics under different behaviors remains a challenging problem. We introduce the multi-choice dynamics model that covers different local dynamics under different behaviors concurrently, which uses different heads to learn the state transition under different behaviors during unsupervised pre-training and selects the most appropriate head for prediction in the downstream task. Experimental results in the manipulation and locomotion domains demonstrate that EUCLID achieves state-of-the-art performance with high sample efficiency, basically solving the state-based URLB benchmark and reaching a mean normalized score of 104.0±1.2% in downstream tasks with 100k fine-tuning steps, which is equivalent to DDPG’s performance at 2M interactive steps with 20× more data. More visualization videos are released on our homepage.

We propose a novel model-fused paradigm for Unsupervised Reinforcement Learning to jointly pre-train the dynamics model and unsupervised exploration policy in the pre-training phase, thus improving the downstream task sampling efficiency.

We are interested in in silico evaluation methodology for molecular optimization methods. Given a sample of molecules and their properties of our interest, we wish not only to train a generator of molecules that can find those optimized with respect to a target property but also to evaluate its performance accurately. A common practice is to train a predictor of the target property on the sample and use it for both training and evaluating the generator. We theoretically investigate this evaluation methodology and show that it potentially suffers from two biases; one is due to misspecification of the predictor and the other to reusing the same sample for training and evaluation. We discuss bias reduction methods for each of the biases, and empirically investigate their effectiveness.

This paper analyzes biases in the evaluation of molecular optimization methods, and methods to alleviate them.

Learning with few labeled tabular samples is often an essential requirement for industrial machine learning applications as varieties of tabular data suffer from high annotation costs or have difficulties in collecting new samples for novel tasks. Despite the utter importance, such a problem is quite under-explored in the field of tabular learning, and existing few-shot learning schemes from other domains are not straightforward to apply, mainly due to the heterogeneous characteristics of tabular data. In this paper, we propose a simple yet effective framework for few-shot semi-supervised tabular learning, coined Self-generated Tasks from UNlabeled Tables (STUNT). Our key idea is to self-generate diverse few-shot tasks by treating randomly chosen columns as a target label. We then employ a meta-learning scheme to learn generalizable knowledge with the constructed tasks. Moreover, we introduce an unsupervised validation scheme for hyperparameter search (and early stopping) by generating a pseudo-validation set using STUNT from unlabeled data. Our experimental results demonstrate that our simple framework brings significant performance gain under various tabular few-shot learning benchmarks, compared to prior semi- and self-supervised baselines. Code is available at https://github.com/jaehyun513/STUNT.

We propose a few-shot tabular learning framework that meta-learns over the self-generated tasks from unlabeled tables.

The study of decentralized learning or independent learning in cooperative multi-agent reinforcement learning has a history of decades. Recently empirical studies show that independent PPO (IPPO) can obtain good performance, close to or even better than the methods of centralized training with decentralized execution, in several benchmarks. However, decentralized actor-critic with convergence guarantee is still open. In this paper, we propose decentralized policy optimization (DPO), a decentralized actor-critic algorithm with monotonic improvement and convergence guarantee. We derive a novel decentralized surrogate for policy optimization such that the monotonic improvement of joint policy can be guaranteed by each agent independently optimizing the surrogate. In practice, this decentralized surrogate can be realized by two adaptive coefficients for policy optimization at each agent. Empirically, we compare DPO with IPPO in a variety of cooperative multi-agent tasks, covering discrete and continuous action spaces, and fully and partially observable environments. The results show DPO outperforms IPPO in most tasks, which can be the evidence for our theoretical results.

A principled algorithm for fully decentralized policy optimization

Inspire by nature world mode, a activation function is proposed. It is absolute function.Through test on mnist dataset and fully-connected neural network and convolutional neural network, some conclusions are put forward. The line of accuracy of absolute function is a little shaken that is different from the line of accuracy of relu and leaky relu. The absolute function can keep the negative parts as equal as the positive parts, so the individualization is more active than relu and leaky relu function. The absolute function is less likely to be over-fitting. Through test on mnist and autoencoder, It is that the leaky relu function can do classification task well, while the absolute function can do generation task well. Because the classification task need more universality and generation task need more individualization. The pleasure irritation and painful irritation is not only the magnitude differences, but also the sign differences, so the negative parts should keep as a part. Stimulation which happens frequently is low value, it is showed around zero in figure 1 . Stimulation which happens accidentally is high value, it is showed far away from zero in figure 1. So the high value is the big stimulation, which is individualization.

A new activition function

Human electroencephalography (EEG) is a brain monitoring modality that senses cortical neuroelectrophysiological activity in high-temporal resolution. One of the greatest challenges posed in applications of EEG is the unstable signal quality susceptible to inevitable artifacts during recordings. To date, most existing techniques for EEG artifact removal and reconstruction are applicable to offline analysis solely, or require individualized training data to facilitate online reconstruction. We have proposed CLEEGN, a novel convolutional neural network for plug-and-play automatic EEG reconstruction. CLEEGN is based on a subject-independent pre-trained model using existing data and can operate on a new user without any further calibration. The performance of CLEEGN was validated using multiple evaluations including waveform observation, reconstruction error assessment, and decoding accuracy on well-studied labeled datasets. The results of simulated online validation suggest that, even without any calibration, CLEEGN can largely preserve inherent brain activity and outperforms leading online/offline artifact removal methods in the decoding accuracy of reconstructed EEG data. In addition, visualization of model parameters and latent features exhibit the model behavior and reveal explainable insights related to existing knowledge of neuroscience. We foresee pervasive applications of CLEEGN in prospective works of online plug-and-play EEG decoding and analysis.

A novel CNN model for training-free online EEG reconstruction with the SOTA performance.

Many Reinforcement Learning tasks rely solely on pixel-based observations of the environment. During deployment, these observations can fall victim to visual perturbations and distortions, causing the agent’s policy to significantly degrade in performance. This motivates the need for robust agents that can generalize in the face of visual distribution shift. One common technique for doing this is to ap- ply augmentations during training; however, it comes at the cost of performance. We propose Augmentation Curriculum Learning a novel curriculum learning ap- proach that schedules augmentation into training into a weak augmentation phase and strong augmentation phase. We also introduce a novel visual augmentation strategy that proves to aid in the benchmarks we evaluate on. Our method achieves state-of-the-art performance on Deep Mind Control Generalization Benchmark.

Combining data augmentation, reinforcement learning and curriculum learning for generalization in reinforcement learning

Federated learning (FL) is an emerging distributed machine learning framework which jointly trains a global model via a large number of local devices with data privacy protections. Its performance suffers from the non-vanishing biases introduced by the local inconsistent optimal and the rugged client-drifts by the local over-fitting. In this paper, we propose a novel and practical method, FedSpeed, to alleviate the negative impacts posed by these problems. Concretely, FedSpeed applies the prox-correction term on the current local updates to efficiently reduce the biases introduced by the prox-term, a necessary regularizer to maintain the strong local consistency. Furthermore, FedSpeed merges the vanilla stochastic gradient with a perturbation computed from an extra gradient ascent step in the neighborhood, thereby alleviating the issue of local over-fitting. Our theoretical analysis indicates that the convergence rate is related to both the communication rounds $T$ and local intervals $K$ with a tighter upper bound $\mathcal{O}(\frac{1}{T})$ if $K=\mathcal{O}(T)$. Moreover, we conduct extensive experiments on the real-world dataset to demonstrate the efficiency of our proposed FedSpeed, which converges significantly faster and achieves the state-of-the-art (SOTA) performance on the general FL experimental settings than several baselines including FedAvg, FedProx, FedCM, FedAdam, SCAFFOLD, FedDyn, FedADMM, etc.

A novel and practical federated learning method with theoretical analysis guarantees achieves higher performance in the common federated settings.

As machine learning becomes more widespread throughout society, aspects including data privacy and fairness must be carefully considered, and are crucial for deployment in highly regulated industries. Unfortunately, the application of privacy enhancing technologies can worsen unfair tendencies in models. In particular, one of the most widely used techniques for private model training, differentially private stochastic gradient descent (DPSGD), frequently intensifies disparate impact on groups within data. In this work we study the fine-grained causes of unfairness in DPSGD and identify gradient misalignment due to inequitable gradient clipping as the most significant source. This observation leads us to a new method for reducing unfairness by preventing gradient misalignment in DPSGD.

DPSGD can have unfair outcomes on protected groups because of direction errors caused by per-sample gradient clipping, but unfairness can be dramatically reduced with a global clipping technique.

Estimating conditional treatment effects has been a longstanding challenge for fields of study such as epidemiology or economics that require a treatment-dosage pair to make decisions, but may not be able to run randomized trials to precisely quantify their effect. In the context of representation learning, there is an extensive literature relating model architectures with regularization techniques to solve this problem using observational data. However, theoretically motivated loss functions and bounds on generalization errors only exist in select circumstances, such as in the presence of binary treatments. In this paper, we introduce new bounds on the counterfactual generalization error in the context of multiple treatments and continuous dosage parameters, which subsume existing results. This result, in a principled manner, guides the definition of new learning objectives that can be used to train representation learning algorithms. We show empirically new state-of-the-art performance results across several benchmark datasets for this problem, including in comparison to doubly-robust estimation methods.

We propose generalization bounds for the counterfactual error in treatment effect estimation in the context of multiple treatments and dosage parameters, and regularization techniques for training prediction models inspired by these bounds.

Visual Parameter-efficient Tuning (VPT) has become a powerful alternative for full fine-tuning, which only updates a small number of parameters while freezing the remaining vast majority of parameters to significantly reduce the storage costs for adapting the pre-trained vision models to downstream tasks. Although the storage burden is largely alleviated, VPT approaches still face many challenges, e.g., lower inference speed and lacking effective configurations for trainable parameters tailored for each task. In this paper, we present a simple yet effective approach termed Sensitivity-aware visual Parameter-efficient Tuning (SPT) to tackle these challenges. Given a desired tunable parameter budget, SPT quickly identifies the important parameters to the given task in a data-dependent way before fine-tuning, without the complex selection schedule. To increase the representational capacity at a negligible cost within the same parameter budget, we employ low-rank reparameterization to achieve a better trade-off between parameter efficiency and accuracy. Through extensive experiments on a wide range of downstream recognition tasks, our SPT achieves better overall transfer performance than the full fine-tuning and the other VPT approaches, with no additional computational or memory overhead during inference. For instance, SPT saves 99.35% of the trainable parameters than the full fine-tuning while achieving a 7.3% higher average top-1 accuracy on VTAB-1k benchmark with the supervised pre-trained ViT-B backbone. Notably, SPT is also the first work that bridges the gap between full fine-tuning and VPT approaches for backbones under self-supervised pre-training strategies MAE and MoCo v3 on the challenging VTAB-1k benchmark.

We propose a visual parameter-efficient tuning approach to identify and tune the parameters at task-specific important positions while being inference-efficient.

Neural networks with sinusoidal activations have been proposed as an alternative to networks with traditional activation functions. Despite their promise, particularly for learning implicit models, their training behavior is not yet fully understood, leading to a number of empirical design choices that are not well justified. In this work, we first propose a simplified version of such sinusoidal neural networks, which allows both for easier practical implementation and simpler theoretical analysis. We then analyze the behavior of these networks from the neural tangent kernel perspective and demonstrate that their kernel approximates a low-pass filter with an adjustable bandwidth. Finally, we utilize these insights to inform the sinusoidal network initialization, optimizing their performance for each of a series of tasks, including learning implicit models and solving differential equations.

We perform a theoretical analysis of a simplified sinusoidal network and use this to propose an informed initialization scheme.

In many task settings, text classification models are likely to encounter examples from novel classes on which they cannot predict correctly. Selective prediction, in which models abstain on low-confidence examples, provides a possible solution, but existing models are often overly confident on OOD examples. To remedy this overconfidence, we introduce Contrastive Novelty Learning (CNL), a two-step method that generates OOD examples representative of novel classes, then trains to decrease confidence on them. First, we generate OOD examples by prompting a large language model twice: we prompt it to enumerate novel classes relevant to the label set, then generate examples from each novel class matching the task format. Second, we train our classifier with a novel contrastive objective that encourages lower confidence on generated OOD examples than training examples. When trained with CNL, classifiers improve in their ability to detect and abstain on OOD examples over prior methods by an average of 2.3% AUAC and 5.5% AUROC across 4 NLP datasets, with no cost to in-distribution accuracy.

We present Contrastive Novelty Learning, a method to improve open-set selective classification by generating probable novel examples with a large language model, then training a classifier for lower relative confidence on generated examples.

No-press Diplomacy is a complex strategy game involving both cooperation and competition that has served as a benchmark for multi-agent AI research. While self-play reinforcement learning has resulted in numerous successes in purely adversarial games like chess, Go, and poker, self-play alone is insufficient for achieving optimal performance in domains involving cooperation with humans. We address this shortcoming by first introducing a planning algorithm we call DiL-piKL that regularizes a reward-maximizing policy toward a human imitation-learned policy. We prove that this is a no-regret learning algorithm under a modified utility function. We then show that DiL-piKL can be extended into a self-play reinforcement learning algorithm we call RL-DiL-piKL that provides a model of human play while simultaneously training an agent that responds well to this human model. We used RL-DiL-piKL to train an agent we name Diplodocus. In a 200-game no-press Diplomacy tournament involving 62 human participants spanning skill levels from beginner to expert, two Diplodocus agents both achieved a higher average score than all other participants who played more than two games, and ranked first and third according to an Elo ratings model.

We train a bot that places first in a no-press Diplomacy tournament with humans by using human-data-regularized reinforcement learning and planning

Deep Neural Networks (DNNs) are prone to learn shortcut patterns that damage the generalization of the DNN during deployment. Shortcut learning is concerning, particularly when the DNNs are applied to safety-critical domains. This paper aims to better understand shortcut learning through the lens of the learning dynamics of the internal neurons during the training process. More specifically, we make the following observations: (1) While previous works treat shortcuts as synonymous with spurious correlations, we emphasize that not all spurious correlations are shortcuts. We show that shortcuts are only those spurious features that are ``easier'' than the core features. (2) We build upon this premise and use instance difficulty methods (like Prediction Depth) to quantify ``easy'' and to identify this behavior during the training phase. (3) We empirically show that shortcut learning can be detected by observing the learning dynamics of the DNN's early layers, irrespective of the network architecture used. In other words, easy features learned by the initial layers of a DNN early during the training are potential shortcuts. We verify our claims on simulated and real medical imaging data and justify the empirical success of our hypothesis by showing the theoretical connections between Prediction Depth and information-theoretic concepts like $\mathcal{V}$-usable information. Lastly, our experiments show the insufficiency of monitoring only accuracy plots during training (as is common in machine learning pipelines), and we highlight the need for monitoring early training dynamics using example difficulty metrics.

Potential shortcuts can be found by monitoring the easy features learned by the initial layers of a DNN early during the training using suitable instance difficulty metrics.

A multitude of work has shown that machine learning-based medical diagnosis systems can be biased against certain subgroups of people. This has motivated a growing number of bias mitigation algorithms that aim to address fairness issues in machine learning. However, it is difficult to compare their effectiveness in medical imaging for two reasons. First, there is little consensus on the criteria to assess fairness. Second, existing bias mitigation algorithms are developed under different settings, e.g., datasets, model selection strategies, backbones, and fairness metrics, making a direct comparison and evaluation based on existing results impossible. In this work, we introduce MEDFAIR, a framework to benchmark the fairness of machine learning models for medical imaging. MEDFAIR covers eleven algorithms from various categories, ten datasets from different imaging modalities, and three model selection criteria. Through extensive experiments, we find that the under-studied issue of model selection criterion can have a significant impact on fairness outcomes; while in contrast, state-of-the-art bias mitigation algorithms do not significantly improve fairness outcomes over empirical risk minimization (ERM) in both in-distribution and out-of-distribution settings. We evaluate fairness from various perspectives and make recommendations for different medical application scenarios that require different ethical principles. Our framework provides a reproducible and easy-to-use entry point for the development and evaluation of future bias mitigation algorithms in deep learning. Code is available at https://github.com/ys-zong/MEDFAIR.

We develop a fairness benchmark for medical imaging and find that the state-of-the-art bias mitigation algorithm does not significantly outperform ERM.

Iterated (a.k.a growing) batch reinforcement learning (RL) is a growing subfield fueled by the demand from systems engineers for intelligent control solutions that they can apply within their technical and organizational constraints. Model-based RL (MBRL) suits this scenario well for its sample efficiency and modularity. Recent MBRL techniques combine efficient neural system models with classical planning (like model predictive control; MPC). In this paper we add two components to this classical setup. The first is a Dyna-style policy learned on the system model using model-free techniques. We call it the guide since it guides the planner. The second component is the explorer, a strategy to expand the limited knowledge of the guide during planning. Through a rigorous ablation study we show that combination of these two ingredients is crucial for optimal performance and better data efficiency. We apply this approach with an off-policy guide and a heating explorer to improve the state of the art of benchmark systems addressing both discrete and continuous action spaces.

Smart agents for resource-limited iterated batch reinforcement learning

Recent works have shown that self-supervised learning can achieve remarkable robustness when integrated with adversarial training (AT). However, the robustness gap between supervised AT (sup-AT) and self-supervised AT (self-AT) remains significant. Motivated by this observation, we revisit existing self-AT methods and discover an inherent dilemma that affects self-AT robustness: either strong or weak data augmentations are harmful to self-AT, and a medium strength is insufficient to bridge the gap. To resolve this dilemma, we propose a simple remedy named DYNACL (Dynamic Adversarial Contrastive Learning). In particular, we propose an augmentation schedule that gradually anneals from a strong augmentation to a weak one to benefit from both extreme cases. Besides, we adopt a fast post-processing stage for adapting it to downstream tasks. Through extensive experiments, we show that DYNACL can improve state-of-the-art self-AT robustness by 8.84% under Auto-Attack on the CIFAR-10 dataset, and can even outperform vanilla supervised adversarial training for the first time. Our code is available at \url{https://github.com/PKU-ML/DYNACL}.

We revisit adversarial contrastive training through the lens of data augmentation, and propose an effective adversarial contrastive framework that outperforms vanilla supervised adversarial robustness.

We study the problem of data subset selection: given a fully labeled dataset and a training procedure, select a subset such that training on that subset yields approximately the same test performance as training on the full dataset. We propose an algorithm, inspired by recent work in machine teaching, that has theoretical guarantees, compelling empirical performance, and is model-agnostic meaning the algorithm's only information comes from the predictions of models trained on subsets. Furthermore, we prove lower bounds that show that our algorithm achieves a subset with near-optimal size (under computational hardness assumptions) while training on a number of subsets that is optimal up to extraneous log factors. We then empirically compare our algorithm, machine teaching algorithms, and coreset techniques on six common image datasets with convolutional neural networks. We find that our machine teaching algorithm can find a subset of CIFAR10 of size less than 16k that yields the same performance (5-6% error) as training on the full dataset of size 50k.

We propose, analyze, and evaluate a machine teaching approach to data subset selection.

There has been an exponential increase in the data generated worldwide. Insights into this data led by machine learning (ML) have given rise to exciting applications such as recommendation engines, conversational agents, and so on. Often, data for these applications is generated at a rate faster than ML pipelines can consume it. In this paper, we propose Density Sketches(DS) - a cheap and practical approach to reducing data redundancy in a streaming fashion. DS creates a succinct online summary of data distribution. While DS does not store the samples from the stream, we can sample unseen data on the fly from DS to use for downstream learning tasks. In this sense, DS can replace actual data in many machine learning pipelines analogous to generative models. Importantly, unlike generative models, which do not have statistical guarantees, the sampling distribution of DS asymptotically converges to underlying unknown density distribution.

online summary of data for density estimation and sampling new data.

Model fairness is an essential element for Trustworthy AI. While many techniques for model fairness have been proposed, most of them assume that the training and deployment data distributions are identical, which is often not true in practice. In particular, when the bias between labels and sensitive groups changes, the fairness of the trained model is directly influenced and can worsen. We make two contributions for solving this problem. First, we analytically show that existing in-processing fair algorithms have fundamental limits in accuracy and fairness. We introduce the notion of correlation shifts, which can explicitly capture the change of the above bias. Second, we propose a novel pre-processing step that samples the input data to reduce correlation shifts and thus enables the in-processing approaches to overcome their limitations. We formulate an optimization problem for adjusting the data ratio among labels and sensitive groups to reflect the shifted correlation. A key advantage of our approach lies in decoupling the roles of pre-processing and in-processing approaches: correlation adjustment via pre-processing and unfairness mitigation on the processed data via in-processing. Experiments show that our framework effectively improves existing in-processing fair algorithms w.r.t. accuracy and fairness, both on synthetic and real datasets.

We analyze fundamental limits in accuracy and fairness of in-processing fair algorithms when the data bias changes with correlation shifts and propose a novel pre-processing step that improves their performances.

Understanding self-supervised learning is important but challenging. Previous theoretical works study the role of pretraining losses, and view neural networks as general black boxes. However, the recent work of [Saunshi et al.] argues that the model architecture --- a component largely ignored by previous works --- also has significant influences on the downstream performance of self-supervised learning. In this work, we provide the first theoretical analysis of self-supervised learning that incorporates the effect of inductive biases originating from the model class. In particular, we focus on contrastive learning --- a popular self-supervised learning method that is widely used in the vision domain. We show that when the model has limited capacity, contrastive representations would recover certain special clustering structures that are compatible with the model architecture, but ignore many other clustering structures in the data distribution. As a result, our theory can capture the more realistic setting where contrastive representations have much lower dimensionality than the number of clusters in the data distribution. We instantiate our theory on several synthetic data distributions, and provide empirical evidence to support the theory.

We provide the first theoretical analysis of self-supervised learning that incorporates the effect of inductive biases of model classes.

We study the problem of training a two-layer neural network (NN) of arbitrary width using stochastic gradient descent (SGD) where the input $\boldsymbol{x}\in \mathbb{R}^d$ is Gaussian and the target $y \in \mathbb{R}$ follows a multiple-index model, i.e., $y=g(\langle\boldsymbol{u_1},\boldsymbol{x}\rangle,...,\langle\boldsymbol{u_k},\boldsymbol{x}\rangle)$ with a noisy link function $g$. We prove that the first-layer weights in the NN converge to the $k$-dimensional principal subspace spanned by the vectors $\boldsymbol{u_1},...,\boldsymbol{u_k}$ of the true model, when online SGD with weight decay is used for training. This phenomenon has several important consequences when $k \ll d$. First, by employing uniform convergence on this smaller subspace, we establish a generalization error bound of $\mathcal{O}(\sqrt{{kd}/{T}})$ after $T$ iterations of SGD, which is independent of the width of the NN. We further demonstrate that, by recovering the principal direction, SGD-trained ReLU NNs can learn a single-index target of the form $y=f(\langle\boldsymbol{u},\boldsymbol{x}\rangle) + \epsilon$ with a sample complexity linear in $d$ (up to log factors), where $f$ is a monotonic function with at most polynomial growth, and $\epsilon$ is the noise. This is in contrast to the known $d^{\Omega(p)}$ samples required to learn any degree $p$ polynomial in the kernel regime, and shows that SGD-trained NNs can outperform the Neural Tangent Kernel at initialization. Finally, we establish compressibility guarantees for NNs using that SGD produces an approximately rank-$k$ first-layer weight matrix.

We prove that SGD on neural networks can learn low-dimensional features in certain settings, and use this to derive novel generalization and excess risk bounds.

Designing effective action spaces for complex environments is a fundamental and challenging problem in reinforcement learning (RL). Some recent works have revealed that naive RL algorithms utilizing well-designed handcrafted discrete action spaces can achieve promising results even when dealing with high-dimensional continuous or hybrid decision-making problems. However, elaborately designing such action spaces requires comprehensive domain knowledge. In this paper, we systemically analyze the advantages of discretization for different action spaces and then propose a unified framework, Neural Discrete Reinforcement Learning (NDRL), to automatically learn how to effectively discretize almost arbitrary action spaces. Specifically, we propose the Action Discretization Variational AutoEncoder (AD-VAE), an action representation learning method that can learn compact latent action spaces while maintain the essential properties of original environments, such as boundary actions and the relationship between different action dimensions. Moreover, we uncover a key issue that parallel optimization of the AD-VAE and online RL agents is often unstable. To address it, we further design several techniques to adapt RL agents to learned action representations, including latent action remapping and ensemble Q-learning. Quantitative experiments and visualization results demonstrate the efficiency and stability of our proposed framework for complex action spaces in various environments.

Discrete all types of action spaces in decision making and utilize arbitrary DRL algorithm to solve them.

Due to the determining role of protein structures on diverse protein functions, pre-training representations of proteins on massive unlabeled protein structures has attracted rising research interests. Among recent efforts on this direction, mutual information (MI) maximization based methods have gained the superiority on various downstream benchmark tasks. The core of these methods is to design correlated views that share common information about a protein. Previous view designs focus on capturing structural motif co-occurrence on the same protein structure, while they cannot capture detailed atom/residue interactions. To address this limitation, we propose the Siamese Diffusion Trajectory Prediction (SiamDiff) method. SiamDiff builds a view as the trajectory that gradually approaches protein native structure from scratch, which facilitates the modeling of atom/residue interactions underlying the protein structural dynamics. Specifically, we employ the multimodal diffusion process as a faithful simulation of the structure-sequence co-diffusion trajectory, where rich patterns of protein structural changes are embedded. On such basis, we design a principled theoretical framework to maximize the MI between correlated multimodal diffusion trajectories. We study the effectiveness of SiamDiff on both residue-level and atom-level structures. On the EC and ATOM3D benchmarks, we extensively compare our method with previous protein structure pre-training approaches. The experimental results verify the consistently superior or competitive performance of SiamDiff on all benchmark tasks compared to existing baselines. The source code will be made public upon acceptance.

In this work, we propose a novel protein structure pre-training algorithm SiamDiff to effectively maximize mutual information between protein structure-sequence co-diffusion trajectories.

In this work, we aim to learn dexterous manipulation of deformable objects using multi-fingered hands. Reinforcement learning approaches for dexterous rigid object manipulation would struggle in this setting due to the complexity of physics interaction with deformable objects. At the same time, previous trajectory optimization approaches with differentiable physics for deformable manipulation would suffer from local optima caused by the explosion of contact modes from hand-object interactions. To address these challenges, we propose DexDeform, a principled framework that abstracts dexterous manipulation skills from human demonstration, and refines the learned skills with differentiable physics. Concretely, we first collect a small set of human demonstrations using teleoperation. And we then train a skill model using demonstrations for planning over action abstractions in imagination. To explore the goal space, we further apply augmentations to the existing deformable shapes in demonstrations and use a gradient optimizer to refine the actions planned by the skill model. Finally, we adopt the refined trajectories as new demonstrations for finetuning the skill model. To evaluate the effectiveness of our approach, we introduce a suite of six challenging dexterous deformable object manipulation tasks. Compared with baselines, DexDeform is able to better explore and generalize across novel goals unseen in the initial human demonstrations. Additional materials can be found at our project website: https://sites.google.com/view/dexdeform.

We investigate the problem of learning dexterous manipulation of deformable objects using multi-fingered hands.

Federated learning (FL) has been a popular research area, where multiple clients collaboratively train a model without sharing their local raw data. Among existing FL solutions, one-shot FL is a promising and challenging direction, where the clients conduct FL training with a single communication round. However, while label skew is a common real-world scenario where some clients may have few or no data of some classes, existing one-shot FL approaches that conduct voting on the local models are not able to produce effective global models. Due to the limited number of classes in each party, the local models misclassify the data from unseen classes into seen classes, which leads to very ineffective global models from voting. To address the label skew issue in one-shot FL, we propose a novel approach named FedOV which generates diverse outliers and introduces them as an additional unknown class in local training to improve the voting performance. Specifically, based on open-set recognition, we propose novel outlier generation approaches by corrupting the original features and further develop adversarial learning to enhance the outliers. Our extensive experiments show that FedOV can significantly improve the test accuracy compared to state-of-the-art approaches in various label skew settings.

We propose FedOV to significantly improve the test accuracy under diverse label skews in one-shot federated learning.

As large language models (LLMs) grow larger and more sophisticated, assessing their "reasoning" capabilities in natural language grows more challenging. Recent question answering (QA) benchmarks that attempt to assess reasoning are often limited by a narrow scope of covered situations and subject matters. We introduce WikiWhy, a QA dataset built around a novel auxiliary task: explaining why an answer is true in natural language. WikiWhy contains over 9,000 "why" question-answer-rationale triples, grounded on Wikipedia facts across a diverse set of topics. Each rationale is a set of supporting statements connecting the question to the answer. WikiWhy serves as a benchmark for the reasoning capabilities of LLMs because it demands rigorous explicit rationales for each answer to demonstrate the acquisition of implicit commonsense knowledge, which is unlikely to be easily memorized. GPT-3 baselines achieve only 38.7% human-evaluated correctness in the end-to-end answer & explain condition, leaving significant room for future improvements.

We propose WikiWhy, a dataset containing 9000+ "why" question-answer-rationale triplets to assess Large Language Models' cause-effect reasoning capability.

A challenging problem for machine learning is few-shot learning, as its models usually require many training samples. Since meta-learning models have strong fine-tuning capabilities for the distribution of tasks, many of them have been applied to few-shot learning. Model-agnostic meta-learning (MAML) is one of the most popular ones. Recent studies showed that MAML-trained models tend to reuse learned features and do not perform strong adaption, especially in the earlier layers. This paper presents an in-detail analysis of this phenomenon by analyzing MAML's components of different variants. Our results show an interesting relationship between the importance of fine-tuning earlier layers and the difference in the distribution between training and testing. As a result, we determine a fundamental weakness of existing MAML variants when the task distribution is heterogeneous, e.g., the numbers of classes do not match during testing and training. We propose a novel nonparametric version of MAML that overcomes these issues while still being able to perform cross-domain adaption.

In our paper, we propose a nonparametric version of MAML which is able to solve problems in heterogeneous enviroments

There has been an growing demand for deep neural network (DNN) powered automatic speech recognition (ASR) on mobile platforms for real-time speech recognition. However, ubiquitous on-device ASR systems are still hindered by two bottlenecks: (1) the lack of large-scale transcribed speech data especially for low-resource spoken languages and (2) the large gap between DNNs' prohibitive complexity and mobiles' limited resources. In parallel, speech models pretrained via self-supervised learning (SSL) have emerged to reduce the reliance on the availability of transcribed speech data, which however further enlarges the efficiency gap because they often adopt large transformers to ensure expressive speech representations. Thus, it is highly desired to trim down the complexity of speech SSL models to enable real-time on-device ASR. This is particularly challenging since only structured sparsity can favor hardware efficiency in commercial devices, under which the speech representation learned by SSL could easily be demolished. To this end, we develop a framework dubbed S$^6$-DAMON to pursue structured sparsity in speech SSL models via data-model co-compression. On the data side, leveraging both the duration of each phoneme and the pauses between the words/phonemes of human utterances, we propose a salient audio token detector, dubbed SALAD, to remove input audio tokens that are redundant; On the model side, we identify that the failure of the SOTA ASR pruning method under structured sparsity is caused by the sparsity discrepancy between finetuning/deployment and their limited learnability of sparsity distributions, and then tackle it via a new ASR pruning pipeline dubbed SAFARI, which adopts a three-step pipeline - sparsify, finetune, and adjust sparsity. Extensive experiments validate that S$^6$-DAMON can enable real-time ASR with limited transcribed speech data requirements while maintaining decent recognition performance. All source codes will be released upon acceptance.

We propose a data-model co-compression framework dubbed S$^6$-DAMON for bridging self-supervised speech models with real-time speech recognition.

Controlling agents remotely with deep reinforcement learning~(DRL) in the real world is yet to come. One crucial stepping stone is to devise RL algorithms that are robust in the face of dropped information from corrupted communication or malfunctioning sensors. Typical RL methods usually require considerable online interaction data that are costly and unsafe to collect in the real world. Furthermore, when applying to the frame dropping scenarios, they perform unsatisfactorily even with moderate drop rates. To address these issues, we propose Decision Transformer under Random Frame Dropping~(DeFog), an offline RL algorithm that enables agents to act robustly in frame dropping scenarios without online interaction. DeFog first randomly masks out data in the offline datasets and explicitly adds the time span of frame dropping as inputs. After that, a finetuning stage on the same offline dataset with a higher mask rate would further boost the performance. Empirical results show that DeFog outperforms strong baselines under severe frame drop rates like 90\%, while maintaining similar returns under non-frame-dropping conditions in the regular MuJoCo control benchmarks and the Atari environments. Our approach offers a robust and deployable solution for controlling agents in real-world environments with limited or unreliable data.

Learning to control against random frame dropping through three original modifications to the Decision Transformer.

Closed-loop optimal control design for high-dimensional nonlinear systems has been a long-standing challenge. Traditional methods, such as solving the associated Hamilton-Jacobi-Bellman equation, suffer from the curse of dimensionality. Recent literature proposed a new promising approach based on supervised learning, by leveraging powerful open-loop optimal control solvers to generate training data and neural networks as efficient high-dimensional function approximators to fit the closed-loop optimal control. This approach successfully handles certain high-dimensional optimal control problems but still performs poorly on more challenging problems. One of the crucial reasons for the failure is the so-called distribution mismatch phenomenon brought by the controlled dynamics. In this paper, we investigate this phenomenon and propose the initial value problem enhanced sampling method to mitigate this problem. We theoretically prove that this sampling strategy improves over the vanilla strategy on the classical linear-quadratic regulator by a factor proportional to the total time duration. We further numerically demonstrate that the proposed sampling strategy significantly improves the performance on tested control problems, including the optimal landing problem of a quadrotor and the optimal reaching problem of a 7 DoF manipulator.

A new adaptive sampling method to improve the performance of the closed-loop controller learned by neural networks

Training agents to control a dynamic environment is a fundamental task in AI. In many environments the dynamics can be summarized by a small set of events that capture the semantic behavior of the system. Typically, these events form chains or cascades. We often wish to change the system behavior using a single intervention that propagates through the cascade. For instance, one may trigger a biochemical cascade to switch the state of a cell, or reroute a truck in logistic chains to meet an unexpected, urgent delivery. We introduce a new supervised learning setup called "Cascade". An agent observes a system with a known dynamics evolving from some initial state. It is given a structured semantic instruction and needs to make an intervention that triggers a cascade of events, such that the system reaches an alternative (counterfactual) behavior. We provide a test-bed for this problem, consisting of physical objects. We combine semantic tree search with an event-driven forward model and devise an algorithm that learns to efficiently search in exponentially large semantic trees of continuous spaces. We demonstrate that our approach learns to effectively follow instructions to intervene in previously unseen complex scenes. When provided an observed cascade of events, it can also reason about alternative outcomes.

We consider the task of controlling the behavior of a cascading process with an intervention at a single point in time. We propose to learn a principled probabilistic scoring function that allows searching efficiently over the space of interventions.

Molecular representation learning (MRL) has gained tremendous attention due to its critical role in learning from limited supervised data for applications like drug design. In most MRL methods, molecules are treated as 1D sequential tokens or 2D topology graphs, limiting their ability to incorporate 3D information for downstream tasks and, in particular, making it almost impossible for 3D geometry prediction/generation. In this paper, we propose a universal 3D MRL framework, called Uni-Mol, that significantly enlarges the representation ability and application scope of MRL schemes. Uni-Mol contains two pretrained models with the same SE(3) Transformer architecture: a molecular model pretrained by 209M molecular conformations; a pocket model pretrained by 3M candidate protein pocket data. Besides, Uni-Mol contains several finetuning strategies to apply the pretrained models to various downstream tasks. By properly incorporating 3D information, Uni-Mol outperforms SOTA in 14/15 molecular property prediction tasks. Moreover, Uni-Mol achieves superior performance in 3D spatial tasks, including protein-ligand binding pose prediction, molecular conformation generation, etc. The code, model, and data are made publicly available at https://github.com/dptech-corp/Uni-Mol.

A universal 3D molecular pretraining framework that significantly enlarges the representation ability and application scope in drug design.

Dropped into an unknown environment, what should an agent do to quickly learn about the environment and how to accomplish diverse tasks within it? We address this question within the goal-conditioned reinforcement learning paradigm, by identifying how the agent should set its goals at training time to maximize exploration. We propose "Planning Exploratory Goals" (PEG), a method that sets goals for each training episode to directly optimize an intrinsic exploration reward. PEG first chooses goal commands such that the agent's goal-conditioned policy, at its current level of training, will end up in states with high exploration potential. It then launches an exploration policy starting at those promising states. To enable this direct optimization, PEG learns world models and adapts sampling-based planning algorithms to "plan goal commands". In challenging simulated robotics environments including a multi-legged ant robot in a maze, and a robot arm on a cluttered tabletop, PEG exploration enables more efficient and effective training of goal-conditioned policies relative to baselines and ablations. Our ant successfully navigates a long maze, and the robot arm successfully builds a stack of three blocks upon command. Website: https://sites.google.com/view/exploratory-goals

We use world models to generate goals for exploration.

Contrastive learning methods train visual encoders by comparing views (e.g., often created via a group of data augmentations on the same instance) from one instance to others. Typically, the views created from one instance are set as positive, while views from other instances are negative. This binary instance discrimination is studied extensively to improve feature representations in self-supervised learning. In this paper, we rethink the instance discrimination framework and find the binary instance labeling insufficient to measure correlations between different samples. For an intuitive example, given a random image instance, there may exist other images in a mini-batch whose content meanings are the same (i.e., belonging to the same category) or partially related (i.e., belonging to a similar category). How to treat the images that correlate similarly to the current image instance leaves an unexplored problem. We thus propose to support the current image by exploring other correlated instances (i.e., soft neighbors). We first carefully cultivate a candidate neighbor set, which will be further utilized to explore the highly-correlated instances. A cross-attention module is then introduced to predict the correlation score (denoted as positiveness) of other correlated instances with respect to the current one. The positiveness score quantitatively measures the positive support from each correlated instance, and is encoded into the objective for pretext training. To this end, our proposed method benefits in discriminating uncorrelated instances while absorbing correlated instances for SSL. We evaluate our soft neighbor contrastive learning method (SNCLR) on standard visual recognition benchmarks, including image classification, object detection, and instance segmentation. The state-of-the-art recognition performance shows that SNCLR is effective in improving feature representations from both ViT and CNN encoders.

We leverage the soft neighbors to sufficiently explore the correlation information among samples in cotrastive learning.

Current point-cloud detection methods have difficulty detecting the open-set objects in the real world, due to their limited generalization capability. Moreover, it is extremely laborious and expensive to collect and fully annotate a point-cloud detection dataset with numerous classes of objects, leading to the limited classes of existing point-cloud datasets and hindering the model to learn general representations to achieve open-set point-cloud detection. Instead of seeking a point-cloud dataset with full labels, we resort to ImageNet1K to broaden the vocabulary of the point-cloud detector. We propose OS-3DETIC, an Open-Set 3D DETector using Image-level Class supervision. Specifically, we take advantage of two modalities, the image modality for recognition and the point-cloud modality for localization, to generate pseudo labels for unseen classes. Then we propose a novel debiased cross-modal cross-task contrastive learning method to transfer the knowledge from image modality to point-cloud modality during training. Without hurting the latency during inference, OS-3DETIC makes the well-known point-cloud detector capable of achieving open-set detection. Extensive experiments demonstrate that the proposed OS-3DETIC achieves at least 10.77 % mAP improvement (absolute value) and 9.56 % mAP improvement (absolute value) by a wide range of baselines on the SUN-RGBD dataset and ScanNet dataset, respectively. Besides, we conduct sufficient experiments to shed light on why the proposed OS-3DETIC works.

We propose an open-set 3D detection method that detects unseen categories without corresponding 3D labels

Models trained via empirical risk minimization (ERM) are known to rely on spurious correlations between labels and task-independent input features, resulting in poor generalization to distributional shifts. Group distributionally robust optimization (G-DRO) can alleviate this problem by minimizing the worst-case loss over a set of pre-defined groups over training data. G-DRO successfully improves performance of the worst group, where the correlation does not hold. However, G-DRO assumes that the spurious correlations and associated worst groups are known in advance, making it challenging to apply them to new tasks with potentially multiple unknown correlations. We propose AGRO---Adversarial Group discovery for Distributionally Robust Optimization---an end-to-end approach that jointly identifies error-prone groups and improves accuracy on them. AGRO equips G-DRO with an adversarial slicing model to find a group assignment for training examples which maximizes worst-case loss over the discovered groups. On the WILDS benchmark, AGRO results in 8\% higher model performance on average on known worst-groups, compared to prior group discovery approaches used with G-DRO. AGRO also improves out-of-distribution performance on SST2, QQP, and MS-COCO---datasets where potential spurious correlations are as yet uncharacterized. Human evaluation of ARGO groups shows that they contain well-defined, yet previously unstudied spurious correlations that lead to model errors.

AGRO is an end-to-end robust optimization technique that discovers error-prone groups and optimizes for their accuracy, resulting in improved robustness to test-time distributional shifts.

Drug recommendation assists doctors in prescribing personalized medications to patients based on their health conditions. Existing drug recommendation solutions adopt the supervised multi-label classification setup and only work with existing drugs with sufficient prescription data from many patients. However, newly approved drugs do not have much historical prescription data and cannot leverage existing drug recommendation methods. To address this, we formulate the new drug recommendation as a few-shot learning problem. Yet, directly applying existing few-shot learning algorithms faces two challenges: (1) complex relations among diseases and drugs and (2) numerous false-negative patients who were eligible but did not yet use the new drugs. To tackle these challenges, we propose EDGE, which can quickly adapt to the recommendation for a new drug with limited prescription data from a few support patients. EDGE maintains a drug-dependent multi-phenotype few-shot learner to bridge the gap between existing and new drugs. Specifically, EDGE leverages the drug ontology to link new drugs to existing drugs with similar treatment effects and learns ontology-based drug representations. Such drug representations are used to customize the metric space of the phenotype-driven patient representations, which are composed of a set of phenotypes capturing complex patient health status. Lastly, EDGE eliminates the false-negative supervision signal using an external drug-disease knowledge base. We evaluate EDGE on two real-world datasets: the public EHR data (MIMIC-IV) and private industrial claims data. Results show that EDGE achieves 7.3% improvement on the ROC-AUC score over the best baseline.

recommendation for new drugs with limited historical prescription data

To afford flexible behaviour, the brain must build internal representations that mirror the structure of variables in the external world. For example, 2D space obeys rules: the same set of actions combine in the same way everywhere (step north, then south, and you won't have moved, wherever you start). We suggest the brain must represent this consistent meaning of actions across space, as it allows you to find new short-cuts and navigate in unfamiliar settings. We term this representation an `actionable representation'. We formulate actionable representations using group and representation theory, and show that, when combined with biological and functional constraints - non-negative firing, bounded neural activity, and precise coding - multiple modules of hexagonal grid cells are the optimal representation of 2D space. We support this claim with intuition, analytic justification, and simulations. Our analytic results normatively explain a set of surprising grid cell phenomena, and make testable predictions for future experiments. Lastly, we highlight the generality of our approach beyond just understanding 2D space. Our work characterises a new principle for understanding and designing flexible internal representations: they should be actionable, allowing animals and machines to predict the consequences of their actions, rather than just encode.

We study a novel definition of an optimal representation of structured spaces, and show that it can be used to derive the brain's grid cells and their perturbations normatively.

Sharpness-Aware Minimization (SAM) has recently emerged as a robust technique for improving the accuracy of deep neural networks. However, SAM incurs a high computational cost in practice, requiring up to twice as much computation as vanilla SGD. The computational challenge posed by SAM arises because each iteration requires both ascent and descent steps and thus double the gradient computations. To address this challenge, we propose to compute gradients in both stages of SAM on only the top-k samples with highest loss. K-SAM is simple and extremely easy-to-implement while providing significant generalization boosts over vanilla SGD at little to no additional cost.

We propose an efficient sharpness-aware minimization by subsampling training data with highest k losses in both gradient calculation steps.

Residual Networks (ResNets) can be interpreted as dynamic systems, which are systems whose state changes over time and can be described with ordinary differential equations (ODEs). Specifically, the dynamic systems interpretation views individual residual blocks as ODEs. The solution to an ODE is an approximation; and therefore contains an error term. If an ODE is stiff it is likely that this error is amplified and becomes dominating in the solution calculations, which negatively affects the accuracy of the approximated solution. Therefore, stiff ODEs are often numerically unstable. In this paper we leverage the dynamic systems interpretation to perform a novel theoretical analysis of ResNets by leveraging findings and tools from numerical analysis of ODEs. Specifically, we perform block level stiffness analysis of ResNets. We find that residual blocks towards the end of ResNet models exhibit increased stiffness and that there is a significant correlation between stiffness and model accuracy and loss. Based on these findings, we propose that ResNets behave as stiff numerically unstable ODEs.

In this paper we are the first ones to connect the concepts of stiffness and ResNets via the dynamical systems interpretation to propose that ResNets can be viewed as stiff ODEs.

While Graph Neural Networks (GNNs) have demonstrated their efficacy in dealing with non-Euclidean structural data, they are difficult to be deployed in real applications due to the scalability constraint imposed by the multi-hop data dependency. Existing methods attempt to address this scalability issue by training student multi-layer perceptrons (MLPs) exclusively on node content features using labels derived from the teacher GNNs. However, the trained MLPs are neither effective nor robust. In this paper, we ascribe the lack of effectiveness and robustness to three significant challenges: 1) the misalignment between content feature and label spaces, 2) the strict hard matching to teacher's output, and 3) the sensitivity to node feature noises. To address the challenges, we propose NOSMOG, a novel method to learn NOise-robust Structure-aware MLPs On Graphs, with remarkable effectiveness, robustness, and efficiency. Specifically, we first address the misalignment by complementing node content with position features to capture the graph structural information. We then design an innovative representational similarity distillation strategy to inject soft node similarities into MLPs. Finally, we introduce adversarial feature augmentation to ensure stable learning against feature noises. Extensive experiments and theoretical analyses demonstrate the superiority of NOSMOG by comparing it to GNNs and the state-of-the-art method in both transductive and inductive settings across seven datasets. Codes are available at https://github.com/meettyj/NOSMOG.

We propose NOSMOG, a novel method to learn noise-robust and structure-aware MLPs on graphs, with superior effectiveness, outstanding robustness, and exceptional efficiency.

We present KeyCLD, a framework to learn Lagrangian dynamics from images. Learned keypoint representations derived from images are directly used as positional state vector for jointly learning constrained Lagrangian dynamics. KeyCLD is trained unsupervised end-to-end on sequences of images. Our method explicitly models the mass matrix, potential energy and the input matrix, thus allowing energy based control. We demonstrate learning of Lagrangian dynamics from images on the dm_control pendulum, cartpole and acrobot environments, wether they are unactuated, underactuated or fully actuated. Trained models are able to produce long-term video predictions, showing that the dynamics are accurately learned. Our method strongly outperforms recent works on learning Lagrangian or Hamiltonian dynamics from images. The benefits of including a Lagrangian prior and prior knowledge of a constraint function is further investigated and empirically evaluated.

We learn unsupervised keypoint representations as state, jointly with constrained Lagrangian dynamics, based on videos of dynamical systems.

Recent breakthroughs in text-to-image synthesis have been driven by diffusion models trained on billions of image-text pairs. Adapting this approach to 3D synthesis would require large-scale datasets of labeled 3D or multiview data and efficient architectures for denoising 3D data, neither of which currently exist. In this work, we circumvent these limitations by using a pretrained 2D text-to-image diffusion model to perform text-to-3D synthesis. We introduce a loss based on probability density distillation that enables the use of a 2D diffusion model as a prior for optimization of a parametric image generator. Using this loss in a DeepDream-like procedure, we optimize a randomly-initialized 3D model (a Neural Radiance Field, or NeRF) via gradient descent such that its 2D renderings from random angles achieve a low loss. The resulting 3D model of the given text can be viewed from any angle, relit by arbitrary illumination, or composited into any 3D environment. Our approach requires no 3D training data and no modifications to the image diffusion model, demonstrating the effectiveness of pretrained image diffusion models as priors.

DeepDream on a pretrained 2D diffusion model enables text-to-3D synthesis

As federated learning (FL) matures, privacy attacks against FL systems in turn become more numerous and complex. Attacks on language models have progressed from recovering single sentences in simple classification tasks to recovering larger parts of user data. Current attacks against federated language models are sequence-agnostic and aim to extract as much data as possible from an FL update - often at the expense of fidelity for any particular sequence. Because of this, current attacks fail to extract any meaningful data under large-scale aggregation. In realistic settings, an attacker cares most about a small portion of user data that contains sensitive personal information, for example sequences containing the phrase "my credit card number is ...". In this work, we propose the first attack on FL that achieves targeted extraction of sequences that contain privacy-critical phrases, whereby we employ maliciously modified parameters to allow the transformer itself to filter relevant sequences from aggregated user data and encode them in the gradient update. Our attack can effectively extract sequences of interest even against extremely large-scale aggregation.

We propose a method that extracts target sequences by keywords under extremely large-scale aggregation in federated learning.

Employing a forward diffusion chain to gradually map the data to a noise distribution, diffusion-based generative models learn how to generate the data by inferring a reverse diffusion chain. However, this approach is slow and costly because it needs many forward and reverse steps. We propose a faster and cheaper approach that adds noise not until the data become pure random noise, but until they reach a hidden noisy data distribution that we can confidently learn. Then, we use fewer reverse steps to generate data by starting from this hidden distribution that is made similar to the noisy data. We reveal that the proposed model can be cast as an adversarial auto-encoder empowered by both the diffusion process and a learnable implicit prior. Experimental results show even with a significantly smaller number of reverse diffusion steps, the proposed truncated diffusion probabilistic models can provide consistent improvements over the non-truncated ones in terms of performance in both unconditional and text-guided image generations.

We propose truncated diffusion probabilistic models, which models an implicit prior to truncate the diffusion chain and requires significantly fewer reverse steps to generate high-quality samples.

Polar codes are widely used state-of-the-art codes for reliable communication that have recently been included in the $5^{\text{th}}$ generation wireless standards ($5$G). However, there still remains room for design of polar decoders that are both efficient and reliable in the short blocklength regime. Motivated by recent successes of data-driven channel decoders, we introduce a novel $\textbf{ C}$ur${\textbf{RI}}$culum based $\textbf{S}$equential neural decoder for $\textbf{P}$olar codes (CRISP). We design a principled curriculum, guided by information-theoretic insights, to train CRISP and show that it outperforms the successive-cancellation (SC) decoder and attains near-optimal reliability performance on the $\text{Polar}(16,32)$ and $\text{Polar}(22,64)$ codes. The choice of the proposed curriculum is critical in achieving the accuracy gains of CRISP, as we show by comparing against other curricula. More notably, CRISP can be readily extended to Polarization-Adjusted-Convolutional (PAC) codes, where existing SC decoders are significantly less reliable. To the best of our knowledge, CRISP constructs the first data-driven decoder for PAC codes and attains near-optimal performance on the $\text{PAC}(16,32)$ code.

We introduce CRISP, a novel curriculum learning based neural decoder that attains near optimal reliability on the Polar code family in the short blocklength regime.

Graph Contrastive learning (GCL) has achieved great success in learning representations from unlabeled graph-structure data. However, how to automatically obtain the optimal contrastive views w.r.t specific downstream tasks is little studied. Theoretically, a downstream task can be causally correlated to particular sub-structures in graphs. The existing GCL methods may fail to enhance model performance on a given task when the task-related semantics are incomplete/preserved in the positive/negative views. To address this problem, we propose G-CENSOR, i.e., Graph Contrastive lEarniNg with taSk-oriented cOunteRfactual views, a model-agnostic framework designed for node property prediction tasks. G-CENSOR can simultaneously generate the optimal task-oriented counterfactual positive/negative views for raw ego-graphs and train graph neural networks (GNNs) with a contrastive objective between the raw ego-graphs and their corresponding counterfac-tual views. Extensive experiments on eight real-world datasets demonstrate that G-CENSOR can consistently outperform existing state-of-the-art GCL methods to improve the task performance and generalizability of a series of typical GNNs. To the best of our knowledge, this is a pioneer investigation to explore task-oriented graph contrastive learning from a counterfactual perspective in node property pre- diction tasks. We will release the source code after the review process.

Graph Contrastive learning with task-oriented counterfactual positive/negative views, a model-agnostic framework designed for node property prediction tasks.

The existing continual learning methods are mainly focused on fully-supervised scenarios and are still not able to take advantage of unlabeled data available in the environment. Some recent works tried to investigate semi-supervised continual learning (SSCL) settings in which the unlabeled data are available, but it is only from the same distribution as the labeled data. This assumption is still not general enough for real-world applications and restricts the utilization of unsupervised data. In this work, we introduce Open-Set Semi-Supervised Continual Learning (OSSCL), a more realistic semi-supervised continual learning setting in which out-of-distribution (OoD) unlabeled samples in the environment are assumed to coexist with the in-distribution ones. Under this configuration, we present a model with two distinct parts: (i) the reference network captures general-purpose and task-agnostic knowledge in the environment by using a broad spectrum of unlabeled samples, (ii) the learner network is designed to learn task-specific representations by exploiting supervised samples. The reference model both provides a pivotal representation space and also segregates unlabeled data to exploit them more efficiently. By performing a diverse range of experiments, we show the superior performance of our model compared with other competitors and prove the effectiveness of each component of the proposed model.

In this paper, we introduced open-set semi-supervised continual learning as a realistic, practical scenario and proposed a novel dual-structured method to perform in this scenario.

Recent studies explored the framing of reinforcement learning as a sequence modeling problem, and then using Transformers to generate effective solutions. In this study, we introduce MCTransformer, a framework that combines Monte-Carlo Tree Search (MCTS) with Transformers. Our approach uses an actor-critic setup, where the MCTS component is responsible for navigating previously-explored states, aided by input from the Transformer. The Transformer controls the exploration and evaluation of new states, enabling an effective and efficient evaluation of various strategies. In addition to the development of highly effective strategies, our setup enables the use of more efficient sampling compared to existing MCTS-based solutions. MCTransformer is therefore able to perform a small number of evaluations for each newly-explored node, and to do so without degrading its performance. Our evaluation, conducted on the challenging and well-known problem of SameGame, shows that our approach outperforms both Transformer-based and MCTS-based solutions.

A novel approach for sequential decision making using reinforcement learning by combining MCTS and transformers.

In many applications of machine learning, like drug discovery and material design, the goal is to generate candidates that simultaneously maximize a set of objectives. As these objectives are often conflicting, there is no single candidate that simultaneously maximizes all objectives, but rather a set of Pareto-optimal candidates where one objective cannot be improved without worsening another. Moreover, these objectives, when considered in practice are often under-specified, making diversity of candidates a key consideration. The existing multi-objective optimization methods focus predominantly on covering the Pareto front, failing the capture diversity in the space of candidates. Motivated by the success of GFlowNets for generation of diverse candidates in a single objective setting, in this paper we consider Multi-Objective GFlowNets (MOGFNs). MOGFNs consist of a Conditional GFlowNet which models a family of single-objective sub-problems derived by decomposing the multi-objective optimization problem. Our work is the first to empirically demonstrate conditional GFlowNets. Through a series of experiments on synthetic tasks and real-world domains, we empirically demonstrate that MOGFNs outperform existing methods in terms of Hypervolume, R2-distance and candidate diversity. We also demonstrate the effectiveness of MOGFNs over existing methods in active learning settings. Finally, we supplement our empirical results with a careful analysis of each component of MOGFNs.

We generate diverse Pareto-optimal candidates for high-dimensional multi-objective optimization problems with GFlowNets.

Standard GPs offer a flexible modelling tool for well-behaved processes. However, deviations from Gaussianity are expected to appear in real world datasets, with structural outliers and shocks routinely observed. In these cases GPs can fail to model uncertainty adequately and may over-smooth inferences. Here we extend the GP framework into a new class of time-changed GPs that allow for straightforward modelling of heavy-tailed non-Gaussian behaviours, while retaining a tractable conditional GP structure through an infinite mixture of non-homogeneous GPs representation. The conditional GP structure is obtained by conditioning the observations on a latent transformed input space and the random evolution of the latent transformation is modelled using a Lévy process which allows Bayesian inference in both the posterior predictive density and the latent transformation function. We present Markov chain Monte Carlo inference procedures for this model and demonstrate the potential benefits compared to a standard GP.

We extend the Gaussian process regression model to allow for locally adaptive behaviour through time-changed GPs and learn latent probabilistic representations of inputs.

In Natural Language Processing (NLP), intelligent neuron models can be susceptible to textual Trojan attacks. Such attacks occur when Trojan models behave normally for standard inputs but generate malicious output for inputs that contain a specific trigger. Syntactic-structure triggers, which are invisible, are becoming more popular for Trojan attacks because they are difficult to detect and defend against. However, these types of attacks require a large corpus of training data to generate poisoned samples with the necessary syntactic structures for Trojan insertion. Obtaining such data can be difficult for attackers, and the process of generating syntactic poisoned triggers and inserting Trojans can be time-consuming. This paper proposes a solution called TrojText, which aims to determine whether invisible textual Trojan attacks can be performed more efficiently and cost-effectively without training data. The proposed approach, called the Representation-Logit Trojan Insertion (RLI) algorithm, uses smaller sampled test data instead of large training data to achieve the desired attack. The paper also introduces two additional techniques, namely the accumulated gradient ranking (AGR) and Trojan Weights Pruning (TWP), to reduce the number of tuned parameters and the attack overhead. The TrojText approach was evaluated on three datasets (AG’s News, SST-2, and OLID) using three NLP models (BERT, XLNet, and DeBERTa). The experiments demonstrated that the TrojText approach achieved a 98.35% classification accuracy for test sentences in the target class on the BERT model for the AG’s News dataset. The source code for TrojText is available at https://github.com/UCF-ML-Research/TrojText.

TrojText is a more realistic, efficient, test-time invisible textual Trojan Insertion method against intelligent neuron models

Random-feature-based attention (RFA) is an efficient approximation of softmax attention with linear runtime and space complexity. However, the approximation gap between RFA and conventional softmax attention is not well studied. Built upon previous progress of RFA, we characterize this gap through the lens of control variates and show that RFA can be decomposed into a sum of multiple control variate estimators for each element in the sequence. This new framework reveals that exact softmax attention can be recovered from RFA by manipulating each control variate. Besides, it allows us to develop a more flexible form of control variates, resulting in a novel attention mechanism that significantly reduces the approximation gap while maintaining linear complexity. Extensive experiments demonstrate that our model outperforms state-of-the-art efficient attention mechanisms on both vision and language tasks.

We present a novel analysis of random feature attention based on control variates, which characterizes its gap to full softmax attention and induces a novel efficient variant that significantly improves the approximation while remaining efficient.

In actor-critic reinforcement learning (RL), the so-called actor and critic, respectively, compute candidate policies and a value function that evaluates the candidate policies. Such RL algorithms may be vulnerable to membership inference attacks (MIAs), a privacy attack that infers the data membership, i.e., whether a specific data record belongs to the training dataset. We investigate the vulnerability of value function in actor-critic to MIAs. We develop \textit{CriticAttack}, a new MIA that targets black-box RL agents by examining the correlation between the expected reward and the value function. We empirically show that \textit{CriticAttack} can correctly infer approximately 90\% of the training data membership, i.e., it achieves 90\% attack accuracy. Such accuracy is far beyond the 50\% random guessing accuracy, indicating a severe privacy vulnerability of the value function. To defend against \textit{CriticAttack}, we design a method called \textit{CriticDefense} that inserts uniform noise to the value function. \textit{CriticDefense} can reduce the attack accuracy to 60\% without significantly affecting the agent’s performance.

We introduce a new membership inference attack focusing on the value function of the actor-critic algorithm.

The method of random Fourier features (RFF), proposed in a seminal paper by Rahimi and Recht (NIPS'07), is a powerful technique to find approximate low-dimensional representations of points in (high-dimensional) kernel space, for shift-invariant kernels. While RFF has been analyzed under various notions of error guarantee, the ability to preserve the kernel distance with \emph{relative} error is less understood. We show that for a significant range of kernels, including the well-known Laplacian kernels, RFF cannot approximate the kernel distance with small relative error using low dimensions. We complement this by showing as long as the shift-invariant kernel is analytic, RFF with $\mathrm{poly}(\epsilon^{-1} \log n)$ dimensions achieves $\epsilon$-relative error for pairwise kernel distance of $n$ points, and the dimension bound is improved to $\mathrm{poly}(\epsilon^{-1}\log k)$ for the specific application of kernel $k$-means. Finally, going beyond RFF, we make the first step towards data-oblivious dimension-reduction for general shift-invariant kernels, and we obtain a similar $\mathrm{poly}(\epsilon^{-1} \log n)$ dimension bound for Laplacian kernels. We also validate the dimension-error tradeoff of our methods on simulated datasets, and they demonstrate superior performance compared with other popular methods including random-projection and Nystr\"{o}m methods.

We characterize for what kernels the random Fourier features method, proposed in a seminal paper by Rahimi and Recht, preserves the relative-error for the kernel distance.

Graph data, such as scene graphs and knowledge graphs, see wide use in AI systems. In real-world and large applications graph data are usually incomplete, motivating graph reasoning models for missing-fact or missing-relationship inference. While these models can achieve state-of-the-art performance, they require a large amount of training data. Recent years have witnessed the rising interest in label creation with data programming (DP) methods, which aim to generate training labels from heuristic labeling functions. However, existing methods typically focus on unstructured data and are not optimized for graphs. In this work, we propose LogicDP, a data programming framework for graph data. Unlike existing DP methods, (1) LogicDP utilizes the inductive logic programming (ILP) technique and automatically discovers the labeling functions from the graph data; (2) LogicDP employs a budget-aware framework to iteratively refine the functions by querying an oracle, which significantly reduces the human efforts in function creations. Experiments show that LogicDP achieves better data efficiency in both scene graph and knowledge graph reasoning tasks.

A data programming framework for generating training labels for graph data

Water impacts the globe daily in new and familiar ways such as the ongoing western United States drought and the 2022 Pakistan flood. These events sustain uncertainty, risk, and loss forces to the global ecosystem. Better forecasting tools are mandatory to calibrate our response in an effort to mitigate such natural hazards in our watersheds and adapt to the planet’s dynamic environment. Here, we present a Deep Convolutional Residual Regressive Neural Net (DCRRNN - pronounced “discern”) platform for obtaining, visualizing, and analyzing the basin response of watersheds to water cycle fluxes. We examine four very large basins, simulating river response to the hydroclimatic fluxes they face. Experiments modulating the lever of time lag between remotely sensed and ground truth measurements are performed to assess the metrological limits of this forecasting device. The resultant grand mean Nash Sutcliffe and Kling Gupta efficiency values are both of greater value than 90\%. Our results show that DCRRNN can become a powerful resource to simulate and forecast the impacts of hydroclimatic events as they relate to watershed response in a globally changing climate.

We analyze and visualize the water cycles of four large US watersheds using representation learning techniques.

Recent work has shown that prompting language models to generate reasoning steps improves performance on many reasoning tasks. When moving beyond prompting, this raises the question of how we should supervise the finetuning of such models: outcome-based approaches which supervise the final result, or process-based approaches which supervise the reasoning process itself? Differences between these approaches might naturally be expected not just in final-answer errors but also in reasoning errors, which can be difficult to detect and are problematic in many real-world domains such as education. We run the first comprehensive comparison between process- and outcome-based approaches trained on a natural language task, GSM8K. We find that pure outcome-based supervision produces similar final-answer error rates with less label supervision. However, for correct reasoning steps we find it necessary to use process-based supervision or supervision from learned reward models that emulate process-based feedback. In total, we improve the previous best results from 16.8% $\rightarrow$ 12.7% final-answer error and 14.0% $\rightarrow$ 3.4% reasoning error among final-answer-correct solutions.

Both process- and outcome-based feedback with all the tricks achieve similar final-answer error rates and SOTA results, but generating accurate reasoning steps requires either process-based supervision, or a reward model that emulates it.

Self-training has been proved an efficient strategy for unsupervised fine-tuning of language models using unlabeled data and model-generated pseudo-labels. However, the performance of self-trained models is unstable under different settings of training and evaluation data, influenced by both data distribution and pseudo-label accuracy. In this work, we propose an outlier robust self-training method based on graduated non-convexity (GNC) to mitigate the problem. We construct self-training as a non-convex optimization problem with outlier training examples. The models are self-trained with robust cost functions based according to Black-Rangarajan Duality. The algorithm learns slack variables as the loss weights for all training samples. The slack variables are used to calibrate the loss items during training to update the model parameters. The calibrated loss items lead to more robust self-trained models against different training and evaluation data and tasks. We conducted experiments on few-shot natural language understanding tasks with labeled and unlabeled data examples. Experiment results show that the proposed loss calibration method improves the performance and stability of self-training under different training tasks and data examples, and also benefits the robustness against adversarial evaluation corpora.

Robust self-trained language understanding against pseudo labeling noises, data imbalance, overfitting, and adversarial evaluation data.

Diversity is a growing research topic in Reinforcement Learning (RL). Previous research on diversity has mainly focused on promoting diversity to encourage exploration and thereby improve quality (the cumulative reward), maximizing diversity subject to quality constraints, or jointly maximizing quality and diversity, known as the quality-diversity problem. In this work, we present the quality-similar diversity problem that features diversity among policies of similar qualities. In contrast to task-agnostic diversity, we focus on task-specific diversity defined by a set of user-specified Behavior Descriptors (BDs). A BD is a scalar function of a trajectory (e.g., the fire action rate for an Atari game), which delivers the type of diversity the user prefers. To derive the gradient of the user-specified diversity with respect to a policy, which is not trivially available, we introduce a set of BD estimators and connect it with the classical policy gradient theorem. Based on the diversity gradient, we develop a population-based RL algorithm to adaptively and efficiently optimize the population diversity at multiple quality levels throughout training. Extensive results on MuJoCo and Atari demonstrate that our algorithm significantly outperforms previous methods in terms of generating user-specified diverse policies across different quality levels.

We formulate the Quality-Similar Diversity (QSD) problem and propose an efficient population-based RL algorithm to optimize the user-defined diversity at multiple quality levels throughout training.

Fully decentralized learning, where the global information, i.e., the actions of other agents, is inaccessible, is a fundamental challenge in multi-agent reinforcement learning. However, the convergence and optimality of most decentralized algorithms are not theoretically guaranteed, since the transition probabilities are non-stationary as all agents are updating policies simultaneously. To tackle this challenge, we propose \textit{best possible operator}, a novel decentralized operator, and prove that the policies of agents will converge to the optimal joint policy if each agent independently updates its individual state-action value by the operator. Further, to make the update more efficient and practical, we simplify the operator and prove that the convergence and optimality still hold with the simplified one. By instantiating the simplified operator, the derived fully decentralized algorithm, best possible Q-learning (BQL), does not suffer from non-stationarity. Empirically, we show that BQL achieves remarkable improvement over baselines in a variety of cooperative multi-agent tasks.

A new Q-learning algorithm for multi-agent reinforcement learning

Several studies have empirically compared in-distribution (ID) and out-of-distribution (OOD) performance of various models. They report frequent positive correlations on benchmarks in computer vision and NLP. Surprisingly, they never observe inverse correlations suggesting necessary trade-offs. This matters to determine whether ID performance can serve as a proxy for OOD generalization. This paper shows that inverse correlations between ID and OOD performance do happen in real-world benchmarks. They could be missed in past studies because of a biased selection of models. We show an example on the WILDS-Camelyon17 dataset, using models from multiple training epochs and random seeds. Our observations are particularly striking with models trained with a regularizer that diversifies the solutions to the ERM objective. We nuance recommendations and conclusions made in past studies. (1) High OOD performance may sometimes require trading off ID performance.(2) Focusing on ID performance alone may not lead to optimal OOD performance: it can lead to diminishing and eventually negative returns in OOD performance. (3) Our example reminds that empirical studies only chart regimes achievable with existing methods: care is warranted in deriving prescriptive recommendations.

Empirical+theoretical examination of an example of inverse correlation between ID/OOD accuracy across multiple neural networks (Camelyon17 dataset).

For machine learning models to be reliable and trustworthy, their decisions must be interpretable. As these models find increasing use in safety-critical applications, it is important that not just the model predictions but also their explanations (as feature attributions) be robust to small human-imperceptible input perturbations. Recent works have shown that many attribution methods are fragile and have proposed improvements in either the attribution methods or the model training. Existing works measure attributional robustness by metrics such as top-$k$ intersection, Spearman's rank-order correlation (or Spearman's $\rho$) or Kendall's rank-order correlation (or Kendall's $\tau$) to quantify the change in feature attributions under input perturbation. However, we show that these metrics are fragile. That is, under such metrics, a simple random perturbation attack can seem to be as significant as more principled attributional attacks. We instead propose Locality-sENSitive (LENS) improvements of the above metrics, namely, LENS-top-$k$, LENS-Spearman and LENS-Kendall, that incorporate the locality of attributions along with their rank order. Our locality-sensitive metrics provide tighter bounds on attributional robustness and do not disproportionately penalize attribution methods for reasonable local changes. We show that the robust attribution methods proposed in recent works also reflect this premise of locality, thus highlighting the need for a locality-sensitive metric for progress in the field. Our empirical results on well-known benchmark datasets using well-known models and attribution methods support our observations and conclusions in this work.

The existing metrics for robustness of attributions to small impercetible perturbations don't capture local sensitivity. We propose new locality-sensitive metrics and show their usefulness.

This paper presents a new Convolutional Neural Network, named Contextual Convolutional Network, that capably serves as a general-purpose backbone for visual recognition. Most existing convolutional backbones follow the representation-to-classification paradigm, where representations of the input are firstly generated by category-agnostic convolutional operations, and then fed into classifiers for specific perceptual tasks (e.g., classification and segmentation). In this paper, we deviate from this classic paradigm and propose to augment potential category memberships as contextual priors in the convolution for contextualized representation learning. Specifically, top-k likely classes from the preceding stage are encoded as a contextual prior vector. Based on this vector and the preceding features, offsets for spatial sampling locations and kernel weights are generated to modulate the convolution operations. The new convolutions can readily replace their plain counterparts in existing CNNs and can be easily trained end-to-end by standard back-propagation without additional supervision. The qualities of Contextual Convolutional Networks make it compatible with a broad range of vision tasks and boost the state-of-the-art architecture ConvNeXt-Tiny by 1.8% on top-1 accuracy of ImageNet classification. The superiority of the proposed model reveals the potential of contextualized representation learning for vision tasks. Code is available at: \url{https://github.com/liang4sx/contextual_cnn}.

In this paper, we propose to augment potential category memberships as contextual priors in the convolution for contextualized representation learning.

Neural network classifiers can largely rely on simple spurious features, such as image backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.

We propose a simple method based on retraining the last layer of a neural network which achieves strong results on spurious correlation benchmarks.

Time-dependent data-generating distributions have proven to be difficult for gradient-based training of neural networks, as the greedy updates result in catastrophic forgetting of previously learned knowledge. Despite the progress in the field of continual learning to overcome this forgetting, we show that a set of common state-of-the-art methods still suffers from substantial forgetting upon starting to learn new tasks, except that this forgetting is temporary and followed by a phase of performance recovery. We refer to this intriguing but potentially problematic phenomenon as the stability gap. The stability gap had likely remained under the radar due to standard practice in the field of evaluating continual learning models only after each task. Instead, we establish a framework for continual evaluation that uses per-iteration evaluation and we define a new set of metrics to quantify worst-case performance. Empirically we show that experience replay, constraint-based replay, knowledge-distillation, and parameter regularization methods are all prone to the stability gap; and that the stability gap can be observed in class-, task-, and domain-incremental learning benchmarks. Additionally, a controlled experiment shows that the stability gap increases when tasks are more dissimilar. Finally, by disentangling gradients into plasticity and stability components, we propose a conceptual explanation for the stability gap.

Proposing an iteration-based continual evaluation framework for CL, we discover, quantify, and analyse the "stability gap", a phenomenon where upon learning new tasks, past tasks exhibit substantial but transient performance loss for SOTA CL methods.

Exploration in environments which differ across episodes has received increasing attention in recent years. Current methods use some combination of global novelty bonuses, computed using the agent's entire training experience, and episodic novelty bonuses, computed using only experience from the current episode. However, the use of these two types of bonuses has been ad-hoc and poorly understood. In this work, we first shed light on the behavior these two kinds of bonuses on hard exploration tasks through easily interpretable examples. We find that the two types of bonuses succeed in different settings, with episodic bonuses being most effective when there is little shared structure between environments and global bonuses being effective when more structure is shared. We also find that combining the two bonuses leads to more robust behavior across both of these settings. Motivated by these findings, we then investigate different algorithmic choices for defining and combining function approximation-based global and episodic bonuses. This results in a new algorithm which sets a new state of the art across 18 tasks from the MiniHack suite used in prior work. Our code is public at \url{web-link}.

We study when episodic and global novelty bonuses are useful in contextual MDPs, and find that it depends on the amount of shared structure across contexts; by combining them, we get SOTA results on MiniHack.

We propose a new Decision Tree Graph Neural Network (DT+GNN) architecture for Graph Neural Network (GNN) explanation. Existing post-hoc explanation methods highlight important inputs but fail to reveal how a GNN uses these inputs. In contrast DT+GNN is fully explainable: Humans can inspect and understand the decision making of DT+GNN at every step. DT+GNN internally uses a novel GNN layer that is restricted to categorical state spaces for nodes and messages. After training with gradient descent we can easily distill these layers into decision trees. These trees are further pruned using our newly proposed method to ensure they are small and easy to interpret. DT+GNN can also compute node-level importance scores like the existing explanation methods. We demonstrate on real-world GNN benchmarks that DT+GNN has competitive classification accuracy and computes competitive explanations. Furthermore, we leverage DT+GNN's full explainability to inspect the decision processes in synthetic and real-world datasets with surprising results. We make this inspection accessible through an interactive web tool.

A new GNN architecture that allows for full explanation not only of the important imputs but also the full decision making process how the inputs are used.

Model-Based Reinforcement Learning (RL) is widely believed to have the potential to improve sample efficiency by allowing an agent to synthesize large amounts of imagined experience. Experience Replay (ER) can be considered a simple kind of model, which has proved extremely effective at improving the stability and efficiency of deep RL. In principle, a learned parametric model could improve on ER by generalizing from real experience to augment the dataset with additional plausible experience. However, owing to the many design choices involved in empirically successful algorithms, it can be very hard to establish where the benefits are actually coming from. Here, we provide theoretical and empirical insight into when, and how, we can expect data generated by a learned model to be useful. First, we provide a general theorem motivating how learning a model as an intermediate step can narrow down the set of possible value functions more than learning a value function directly from data using the Bellman equation. Second, we provide an illustrative example showing empirically how a similar effect occurs in a more concrete setting with neural network function approximation. Finally, we provide extensive experiments showing the benefit of model-based learning for online RL in environments with combinatorial complexity, but factored structure that allows a learned model to generalize. In these experiments, we take care to control for other factors in order to isolate, insofar as possible, the benefit of using experience generated by a learned model relative to ER alone.

We show how algorithms that generate experience with a learned parametric model can generalize in a way that is inherently more powerful than relying on value-function generalization and experience replay.

We study reinforcement learning (RL) with linear function approximation, unknown transition, and adversarial losses in the bandit feedback setting. Specifically, the unknown transition probability function is a linear mixture model \citep{AyoubJSWY20,ZhouGS21,HeZG22} with a given feature mapping, and the learner only observes the losses of the experienced state-action pairs instead of the whole loss function. We propose an efficient algorithm LSUOB-REPS which achieves $\widetilde{O}(dS^2\sqrt{K}+\sqrt{HSAK})$ regret guarantee with high probability, where $d$ is the ambient dimension of the feature mapping, $S$ is the size of the state space, $A$ is the size of the action space, $H$ is the episode length and $K$ is the number of episodes. Furthermore, we also prove a lower bound of order $\Omega(dH\sqrt{K}+\sqrt{HSAK})$ for this setting. To the best of our knowledge, we make the first step to establish a provably efficient algorithm with a sublinear regret guarantee in this challenging setting and solve the open problem of \citet{HeZG22}.

We make the first step to establish a provably efficient algorithm in adversarial linear mixture mdps with bandit feedback and unknown transition.

In this paper, we provide a deep analysis of temporal modeling for action recognition, an important but underexplored problem in the literature. We first propose a new approach to quantify the temporal relationships between frames captured by CNN-based action models based on layer-wise relevance propagation. We then conduct comprehensive experiments and in-depth analysis to provide a better understanding of how temporal modeling is affected by various factors such as dataset, network architecture, and input frames. With this, we further study some important questions for action recognition that lead to interesting findings. Our analysis shows that there is no strong correlation between temporal relevance and model performance; and action models tend to capture local temporal information, but less long-range dependencies.

The paper provides a deep analysis of the temporal modeling for action recognition.

We study the task of training regression models with the guarantee of _label_ differential privacy (DP). Based on a global prior distribution of label values, which could be obtained privately, we derive a label DP randomization mechanism that is optimal under a given regression loss function. We prove that the optimal mechanism takes the form of a "randomized response on bins", and propose an efficient algorithm for finding the optimal bin values. We carry out a thorough experimental evaluation on several datasets demonstrating the efficacy of our algorithm.

We present a new label differentially private algorithm for training regression models.

This paper novelly breaks down with ignorable loss an RNN layer into a sequence of simple RNNs, each of which can be further rewritten into a lightweight positional encoding matrix of a self-attention, named the Recurrence Encoding Matrix (REM). Thus, recurrent dynamics introduced by the RNN layer can be encapsulated into the positional encodings of a multihead self-attention, and this makes it possible to seamlessly incorporate these recurrent dynamics into a Transformer, leading to a new module, Self-Attention with Recurrence (RSA). The proposed module can leverage the recurrent inductive bias of REMs to achieve a better sample efficiency than its corresponding baseline Transformer, while the self-attention is used to model the remaining non-recurrent signals. The relative proportions of these two components are controlled by a data-driven gated mechanism, and the effectiveness of RSA modules are demonstrated by four sequential learning tasks.

We propose a new module to encode the recurrent dynamics of an RNN layer into Transformers and higher sample efficiency can be achieved.

There has been great progress on equilibrium-finding research over the last 20 years. Most of that work has focused on games with finite, discrete action spaces. However, many games involving space, time, money, etc. have continuous action spaces. We study the problem of computing approximate Nash equilibria of games with continuous strategy sets. The main measure of closeness to Nash equilibrium is exploitability, which measures how much players can benefit from unilaterally changing their strategy. We propose a new method that minimizes an approximation of exploitability with respect to the strategy profile. This approximation is computed using learned best-response functions, which take the current strategy profile as input and return learned best responses. The strategy profile and best-response functions are trained simultaneously, with the former trying to minimize exploitability while the latter try to maximize it. We evaluate our method on various continuous games, showing that it outperforms prior methods.

We propose a new method for equilibrium finding based on the idea of learned best-response functions.

Group imbalance has been a known problem in empirical risk minimization (ERM), where the achieved high \textit{average} accuracy could be accompanied by low accuracy in a \textit{minority} group. Despite various algorithmic efforts to improve the minority group accuracy, a theoretical study of the generalization performance of ERM on individual groups remains elusive. By formulating the group imbalance problem with the Gaussian Mixture Model, this paper quantifies the impact of individual groups on the sample complexity, the convergence rate, and the average and group-level testing performance. Although our theoretical framework is centered on binary classification using a one-hidden-layer neural network, to the best of our knowledge, we provide the first theoretical analysis of the group-level generalization of ERM in addition to the commonly studied average generalization performance. Sample insights of our theoretical results include that when all group-level co-variance is in the medium regime and all mean are close to zero, the learning performance is most desirable in the sense of a small sample complexity, a fast training rate, and a high average and group-level testing accuracy. Moreover, we show that increasing the fraction of the minority group in the training data does not necessarily improve the generalization performance of the minority group. Our theoretical results are validated on both synthetic and empirical datasets such as CelebA and CIFAR-10 in image classification.

A theoretical characterization of generalization and sample complexity of training neural networks with group imbalance

Multi-input multi-output training improves network performance by optimizing multiple subnetworks simultaneously. In this paper, we propose MixViT, the first MIMO framework for vision transformers that takes advantage of ViTs’ innate mechanisms to share features between subnetworks. This is in stark contrast to traditional MIMO CNNs that are limited by their inability to mutualize features. Unlike them, MixViT only separates subnetworks in the last layers thanks to a novel source attribution that ties tokens to specific subnetworks. As such, we retain the benefits of multi-output supervision while training strong features useful to both subnetworks. We verify MixViT leads to significant gains across multiple architectures (ConViT, CaiT) and datasets (CIFAR, TinyImageNet, ImageNet-100, and ImageNet-1k) by fitting multiple subnetworks at the end of a base model.

We propose MixViT, an inexpensive framework that improves vision transformers by training multiple subnetworks at the end of the model through multi-input multi-output training.

Accurate time series forecasting is a fundamental challenge in data science. Unlike traditional statistical methods, conventional machine learning models, such as RNNs and CNNs, use historical data consisting of previously measured variables including the forecast variable and all its covariates. However, in many applications, some of the covariates can be predicted with reasonable accuracy for the immediate future. Note that the input may also contain some covariates that cannot be accurately predicted. We consider the problem of predicting water levels at a given location in a river or canal system using historical data and future covariates, some of which (e.g., precipitation, tide) may be predictable. In many applications, for some covariates of interest, it may be possible to use historical data or accurate predictions for the near future. Traditional methods to incorporate future predictable covariates have major limitations. The strategy of simply concatenating the future predicted covariates to the input vector is highly likely to miss the past-future connection. Another strategy that iteratively predicts one step at a time can end up with prediction error accumulation. We propose two novel feature representation strategies to solve those limitations -- shifting and padding, which create a framework for contextually linking the past with the predicted future, while avoiding any accumulation of prediction errors. Extensive experiments on three well-known datasets revealed that our strategies when applied to RNN and CNN backbones, outperform existing methods. Our experiments also suggest a relationship between the amount of shifting and padding and the periodicity of the time series.

We propose two novel feature representation strategies for time series forecasting with predicted future covariates.

Deep Active Learning (DAL) has been advocated as a promising method to reduce labeling costs in supervised learning. However, existing evaluations of DAL methods are based on different settings, and their results are controversial. To tackle this issue, this paper comprehensively evaluates 19 existing DAL methods in a uniform setting, including traditional fully-\underline{s}upervised \underline{a}ctive \underline{l}earning (SAL) strategies and emerging \underline{s}emi-\underline{s}upervised \underline{a}ctive \underline{l}earning (SSAL) techniques. We have several non-trivial findings. First, most SAL methods cannot achieve higher accuracy than random selection. Second, semi-supervised training brings significant performance improvement compared to pure SAL methods. Third, performing data selection in the SSAL setting can achieve a significant and consistent performance improvement, especially with abundant unlabeled data. Our findings produce the following guidance for practitioners: one should (i) apply SSAL as early as possible and (ii) collect more unlabeled data whenever possible, for better model performance. We will release our code upon acceptance.

Our paper provides a comprehensive empirical study of existing deep active learning methods.

We investigate the ability of language models to perform compositional reasoning tasks where the overall solution depends on correctly composing the answers to sub-problems. We measure how often models can correctly answer all sub-problems but not generate the overall solution, a ratio we call the compositionality gap. We evaluate this ratio by asking multi-hop questions with answers that require composing multiple facts unlikely to have been observed together during pretraining. In the GPT-3 family of models, as model size increases we show that the single-hop question answering performance improves faster than the multi-hop performance does, therefore the compositionality gap does not decrease. This surprising result suggests that while more powerful models memorize and recall more factual knowledge, they show no corresponding improvement in their ability to perform this kind of compositional reasoning. We then demonstrate how elicitive prompting (such as chain of thought) narrows the compositionality gap by reasoning explicitly instead of implicitly. We present a new method, self-ask, that further improves on chain of thought. In our method, the model explicitly asks itself (and then answers) follow-up questions before answering the initial question. We finally show that self-ask's structured prompting lets us easily plug in a search engine to answer the follow-up questions, which additionally improves accuracy.

Language models can solve complex problems when you let them 'talk things through', our new 'self-ask' prompt improves their ability to do that, and lets us easily plug in a search engine for even better performance.

In this paper, we first extend the recent Masked Auto-Encoder (MAE) model from a single modality to audio-visual multi-modalities. Subsequently, we propose the Contrastive Audio-Visual Masked Auto-Encoder (CAV-MAE) by combining contrastive learning and masked data modeling, two major self-supervised learning frameworks, to learn a joint and coordinated audio-visual representation. Our experiments show that the contrastive audio-visual correspondence learning objective not only enables the model to perform audio-visual retrieval tasks, but also helps the model learn a better joint representation. As a result, our fully self-supervised pretrained CAV-MAE achieves a new SOTA accuracy of 65.9% on VGGSound, and is comparable with the previous best supervised pretrained model on AudioSet in the audio-visual event classification task. Code and pretrained models are at https://github.com/yuangongnd/cav-mae.

We propose the Contrastive Audio-Visual Masked Auto-Encoder that combines contrastive learning and masked data modeling, two major self-supervised learning frameworks, to learn a joint and coordinated audio-visual representation.

It has been observed in practice that applying pruning-at-initialization methods to neural networks and training the sparsified networks can not only retain the testing performance of the original dense models, but also sometimes even slightly boost the generalization performance. Theoretical understanding for such experimental observations are yet to be developed. This work makes the first attempt to study how different pruning fractions affect the model's gradient descent dynamics and generalization. Specifically, this work considers a classification task for overparameterized two-layer neural networks, where the network is randomly pruned according to different rates at the initialization. It is shown that as long as the pruning fraction is below a certain threshold, gradient descent can drive the training loss toward zero and the network exhibits good generalization performance. More surprisingly, the generalization bound gets better as the pruning fraction gets larger. To complement this positive result, this work further shows a negative result: there exists a large pruning fraction such that while gradient descent is still able to drive the training loss toward zero (by memorizing noise), the generalization performance is no better than random guessing. This further suggests that pruning can change the feature learning process, which leads to the performance drop of the pruned neural network. Up to our knowledge, this is the first generalization result for pruned neural networks, suggesting that pruning can improve the neural network's generalization.

We study the effect of pruning under different rates on neural network generalization.

Neural networks suffer from catastrophic forgetting when sequentially learning tasks phase-by-phase, making them inapplicable in dynamically updated systems. Class-incremental learning (CIL) aims to enable neural networks to learn different categories at multi-stages. Recently, dynamic-structure-based CIL methods achieve remarkable performance. However, these methods train all modules in a coupled manner and do not consider possible conflicts among modules, resulting in spoilage of eventual predictions. In this work, we propose a unifying energy-based theory and framework called Bi-Compatible Energy-Based Expansion and Fusion (BEEF) to analyze and achieve the goal of CIL. We demonstrate the possibility of training independent modules in a decoupled manner while achieving bi-directional compatibility among modules through two additionally allocated prototypes, and then integrating them into a unifying classifier with minimal cost. Furthermore, BEEF extends the exemplar-set to a more challenging setting, where exemplars are randomly selected and imbalanced, and maintains its performance when prior methods fail dramatically. Extensive experiments on three widely used benchmarks: CIFAR-100, ImageNet-100, and ImageNet-1000 demonstrate that BEEF achieves state-of-the-art performance in both the ordinary and challenging CIL settings. The Code is available at https://github.com/G-U-N/ICLR23-BEEF.

A unifying energy-based theory and framework called 3EF to analyze and achieve the goal of class-incremental learning.

Recent studies of learning algorithms have shown that there is a regime with an initial increase in the largest eigenvalue of the loss Hessian (progressive sharpening), followed by a stabilization of the eigenvalue near the maximum value which allows convergence (edge of stability). We consider a class of predictive models that are quadratic in the parameters, which we call second-order regression models. This is in contrast with the neural tangent kernel regime, where the predictive function is linear in the parameters. For quadratic objectives in two dimensions, we prove that this second order regression model exhibits both progressive sharpening and edge of stability behavior. We then show that in higher dimensions, the model shows this behavior generically without the structure of a neural network, due to a non-linearity induced in the learning dynamics. Finally, we show that edge of stability behavior in neural networks is correlated with the behavior in quadratic regression models.

Recently observed non-linear learning effects like progressive sharpening and edge of stability occur generically in a simple, second order regression model.

Masked image modeling (MIM) has become a popular strategy for self-supervised learning (SSL) of visual representations with Vision Transformers. A representative MIM model, the masked auto-encoder (MAE), randomly masks a subset of image patches and reconstructs the masked patches given the unmasked patches. Concurrently, many recent works in self-supervised learning utilize the student/teacher paradigm which provides the student with an additional target based on the output of a teacher composed of an exponential moving average (EMA) of previous students. Although common, relatively little is known about the dynamics of the interaction between the student and teacher. Through analysis on a simple linear model, we find that the teacher conditionally removes previous gradient directions based on feature similarities which effectively acts as a conditional momentum regularizer. From this analysis, we present a simple SSL method, the Reconstruction-Consistent Masked Auto-Encoder (RC-MAE) by adding an EMA teacher to MAE. We find that RC-MAE converges faster and requires less memory usage than state-of-the-art self-distillation methods during pre-training, which may provide a way to enhance the practicality of prohibitively expensive self-supervised learning of Vision Transformer models. Additionally, we show that RC-MAE achieves more robustness and better performance compared to MAE on downstream tasks such as ImageNet-1K classification, object detection, and instance segmentation.

We conduct analysis of the dynamics of the self-distillation scheme in masked auto-encoder.

Despite exciting progress in large-scale language generation, the expressiveness of its representations is severely limited by the \textit{anisotropy} issue where the hidden representations are distributed into a narrow cone in the vector space. To address this issue, we present ContraGen, a novel contrastive learning framework to improve the representation with better uniformity and discrimination at both sequence-level and token-level. We assess ContraGen on a wide range of downstream tasks in natural and programming languages. We show that ContraGen can effectively enhance both uniformity and discrimination of the representations and lead to the desired improvement on various language understanding tasks where discriminative representations are crucial for attaining good performance. Specifically, we attain $45.9\%$ relative improvement on the Semantic Textual Similarity tasks and $33.5\%$ on Code-to-Code Search tasks. Furthermore, by improving the expressiveness of the representations, ContraGen also boosts the source code generation capability with $9\%$ relative improvement on execution accuracy on HumanEval benchmark.

An effective contrastive learning approach that enhances both the generation and discrimination capability of causal language models

Diffusion models have become competitive candidates for solving various inverse problems. Models trained for specific inverse problems work well but are limited to their particular use cases, whereas methods that use problem-agnostic models are general but often perform worse empirically. To address this dilemma, we introduce Pseudoinverse-guided Diffusion Models ($\Pi$GDM), an approach that uses problem-agnostic models to close the gap in performance. $\Pi$GDM directly estimates conditional scores from the measurement model of the inverse problem without additional training. It can address inverse problems with noisy, non-linear, or even non-differentiable measurements, in contrast to many existing approaches that are limited to noiseless linear ones. We illustrate the empirical effectiveness of $\Pi$GDM on several image restoration tasks, including super-resolution, inpainting and JPEG restoration. On ImageNet, $\Pi$GDM is competitive with state-of-the-art diffusion models trained on specific tasks, and is the first to achieve this with problem-agnostic diffusion models. $\Pi$GDM can also solve a wider set of inverse problems where the measurement processes are composed of several simpler ones.

We introduce pseudoinverse guidance, an approach to solve inverse problems with generative diffusion models.

Symbolic knowledge graphs (KGs) have been constructed either by expensive human crowdsourcing or with complex text mining pipelines. The emerging large pretrained language models (LMs), such as BERT, have shown to implicitly encode massive knowledge which can be queried with properly designed prompts. However, compared to the explicit KGs, the implict knowledge in the black-box LMs is often difficult to access or edit and lacks explainability. In this work, we aim at harvesting symbolic KGs from the LMs, and propose a new framework for automatic KG construction empowered by the neural LMs’ flexibility and scalability. Compared to prior works that often rely on large human annotated data or existing massive KGs, our approach requires only the minimal definition of relations as inputs, and hence is suitable for extracting knowledge of rich new relations that are instantly assigned and not available before. The framework automatically generates diverse prompts, and performs efficient knowledge search within a given LM for consistent outputs. The knowledge harvested with our approach shows competitive quality, diversity, and novelty. As a result, we derive from diverse LMs a family of new KGs (e.g., BERTNET and ROBERTANET) that contain a richer set of relations, including some complex ones (e.g., "A is capable of but not good at B") that cannot be extracted with previous methods. Besides, the resulting KGs also serve as a vehicle to interpret the respective source LMs, leading to new insights into the varying knowledge capability of different LMs.

In this work, we aim at harvesting symbolic KGs from the LMs, and propose a new framework for automatic KG construction empowered by the neural LMs' flexibility and scalability.

Gradient Descent (GD) is a powerful workhorse of modern machine learning thanks to its scalability and efficiency in high-dimensional spaces. Its ability to find local minimisers is only guaranteed for losses with Lipschitz gradients, where it can be seen as a `bona-fide' discretisation of an underlying gradient flow. Yet, many ML setups involving overparametrised models do not fall into this problem class, which has motivated research beyond the so-called ``Edge of Stability'' (EoS), where the step-size crosses the admissibility threshold inversely proportional to the Lipschitz constant above. Perhaps surprisingly, GD has been empirically observed to still converge regardless of local instability and oscillatory behavior. The incipient theoretical analysis of this phenomena has mainly focused in the overparametrised regime, where the effect of choosing a large learning rate may be associated to a `Sharpness-Minimisation' implicit regularisation within the manifold of minimisers, under appropriate asymptotic limits. In contrast, in this work we directly examine the conditions for such unstable convergence, focusing on simple, yet representative, learning problems. Specifically, we characterize a local condition involving third-order derivatives that stabilizes oscillations of GD above the EoS, and leverage such property in a teacher-student setting, under population loss. Finally, focusing on Matrix Factorization, we establish a non-asymptotic `Local Implicit Bias' of GD above the EoS, whereby quasi-symmetric initializations converge to symmetric solutions --- where sharpness is minimum amongst all minimisers.

We prove convergence results of Gradient Descent beyond Edge of Stability in several nonlinear and high-dimensional problems.

We introduce the novel OCL-PDS problem - Online Continual Learning for Progressive Distribution Shift. PDS refers to the subtle, gradual, and continuous distribution shift that widely exists in modern deep learning applications. It is widely observed in industry that PDS can cause significant performance drop. While previous work in continual learning and domain adaptation addresses this problem to some extent, our investigations from the practitioner's perspective reveal flawed assumptions that limit their applicability on daily challenges faced in real-world scenarios, and this work aims to close the gap between academic research and industry. For this new problem, we build 4 new benchmarks from the Wilds dataset, and implement 12 algorithms and baselines including both supervised and semi-supervised methods, which we test extensively on the new benchmarks. We hope that this work can provide practitioners with tools to better handle realistic PDS, and help scientists design better OCL algorithms.

We introduce the novel OCL-PDS problem for studying gradual distribution shift with time, and release 4 new benchmarks and 12 algorithms/baselines implementation for this new problem.

The graph neural network (GNN) has presented impressive achievements in numerous machine learning tasks. However, many existing GNN models are shown to be extremely vulnerable to adversarial attacks, which makes it essential to build robust GNN architectures. In this work, we propose a novel interpretable message passing scheme with adaptive structure (ASMP) to defend against adversarial attacks on graph structure. Layers in ASMP are derived based on optimization steps that minimize an objective function that simultaneously learns the node feature and the graph structure. ASMP is adaptive in the sense that the message passing process in different layers is able to be carried out over different graphs. Such a property allows more fine-grained handling of the noisy graph structure and hence improves the robustness. Integrating ASMP with neural networks can lead to a new family of GNNs with adaptive structure (ASGNN). Extensive experiments on semi-supervised node classification tasks demonstrate that the proposed ASGNN outperforms the state-of-the-art GNN architectures with respect to classification performance under various graph adversarial attacks.

A novel graph neural network model with adaptive structure that has strong resilience to graph structural attacks

Recently, efforts have been made in the community to design new Graph Neural Networks (GNN), as limitations of Message Passing Neural Networks became more apparent. This led to the appearance of Graph Transformers using global graph features such as Laplacian Eigenmaps. In our paper, we introduce a GNN architecture where the aggregation weights are computed using the long-range correlations of a quantum system. These correlations are generated by translating the graph topology into the interactions of a set of qubits in a quantum computer. The recent development of quantum processing units enables the computation of a new family of global graph features that would be otherwise out of reach for classical hardware. We give some theoretical insights about the potential benefits of this approach, and benchmark our algorithm on standard datasets. Although not being adapted to all datasets, our model performs similarly to standard GNN architectures, and paves a promising future for quantum enhanced GNNs.

A new Graph Neural Network architecture where the aggregation weights are computed with a quantum computer.

Text-to-image (T2I) models enable controllable image generation through user-provided captions. A text encoder is typically used to map captions to a latent space, and it has been shown to be critical for model's performance. However, replacing or upgrading the text encoder in a T2I model is challenging due to the tight bond between the current encoder and the image decoder. It requires training the model from scratch, which can be prohibitively expensive. To address this problem, we introduce a more efficient approach to align a pre-trained language model with the latent space of an existing T2I model. We propose a Model Translation Network (MTN) and a new training objective to align the representation spaces of the two text encoders using only a corpus of unlabeled text. We empirically find that MTN can be trained efficiently and can boost the performance of existing T2I models by upgrading their text encoder. Moreover, we find that MTN can align multilingual language models such as XLM-Roberta, thus allowing existing T2I models to generate high-quality images from captions beyond English.

This paper introduces a new method to efficiently plug new language models to exiting text-to-image generation models as enhancement in scalability.

Intrinsic motivation is a critical ingredient in reinforcement learning to enable progress when rewards are sparse. However, many existing approaches that measure the novelty of observations are brittle, or rely on restrictive assumptions about the environment which limit generality. We propose to decompose the exploration problem into two orthogonal sub-problems: (i) finding the right representation (metric) for exploration (ii) estimating densities in this representation space. To address (ii), we introduce Robust Exploration via Clustering-based Online Density Estimation (RECODE), a non-parametric method that estimates visitation counts for clusters of states that are similar according to the metric induced by any arbitrary representation learning technique. We adapt classical clustering algorithms to design a new type of memory that allows RECODE to keep track of the history of interactions over thousands of episodes, thus effectively tracking global visitation counts. This is in contrast to existing non-parametric approaches, that can only store the recent history, typically the current episode. The generality of RECODE allows us to easily address (i) by leveraging both off-the-shelf and novel representation learning techniques. In particular, we introduce a novel generalization of the action-prediction representation that leverages multi-step predictions and that we find to be better suited to a suite of challenging 3D-exploration tasks in DM-HARD-8. We show experimentally that our approach can work with a variety of RL agents, and obtain state-of-the-art performance on Atari and DM-HARD-8.

We derive a theoretically sound and effective exploration bonus for deep RL using online clustering to estimate visitation densities in a learned latent representation space

Continual learning is a problem for artificial neural networks that their biological counterparts are adept at solving. Building on work using Sparse Distributed Memory (SDM) to connect a core neural circuit with the powerful Transformer model, we create a modified Multi-Layered Perceptron (MLP) that is a strong continual learner. We find that every component of our MLP variant translated from biology is necessary for continual learning. Our solution is also free from any memory replay or task information, and introduces novel methods to train sparse networks that may be broadly applicable.

Improving Sparse Distributed Memory via additional neurobiology results in a deep learning model with strong, organic continual learning and insights into sparse models more broadly.

MuZero Unplugged presents a promising approach for offline policy learning from logged data. It conducts Monte-Carlo Tree Search (MCTS) with a learned model and leverages Reanalyze algorithm to learn purely from offline data. For good performance, MCTS requires accurate learned models and a large number of simulations, thus costing huge computing time. This paper investigates a few hypotheses where MuZero Unplugged may not work well under the offline RL settings, including 1) learning with limited data coverage; 2) learning from offline data of stochastic environments; 3) improperly parameterized models given the offline data; 4) with a low compute budget. We propose to use a regularized one-step look-ahead approach to tackle the above issues. Instead of planning with the expensive MCTS, we use the learned model to construct an advantage estimation based on a one-step rollout. Policy improvements are towards the direction that maximizes the estimated advantage with regularization of the dataset. We conduct extensive empirical studies with BSuite environments to verify the hypotheses and then run our algorithm on the RL Unplugged Atari benchmark. Experimental results show that our proposed approach achieves stable performance even with an inaccurate learned model. On the large-scale Atari benchmark, the proposed method outperforms MuZero Unplugged by 43%. Most significantly, it uses only 5.6% wall-clock time (i.e., 1 hour) compared to MuZero Unplugged (i.e., 17.8 hours) to achieve a 150% IQM normalized score with the same hardware and software stacks. Our implementation is open-sourced at https://github.com/sail-sg/rosmo.

We propose a regularized one-step model-based method that outperforms MuZero Unplugged on Atari benchmark.

In this paper, we revisit the problem of differentially private empirical risk minimization (DP-ERM) and differentially private stochastic convex optimization (DP-SCO). We show that a well-studied continuous time algorithm from statistical physics, called Langevin diffusion (LD), simultaneously provides optimal privacy/utility trade-offs for both DP-ERM and DP-SCO, under $\epsilon$-DP, and $(\epsilon,\delta)$-DP both for convex and strongly convex loss functions. We provide new time and dimension independent uniform stability properties of LD, with which we provide the corresponding optimal excess population risk guarantees for $\epsilon$-DP. An important attribute of our DP-SCO guarantees for $\epsilon$-DP is that they match the non-private optimal bounds as $\epsilon\to\infty$.

We show that Langevin diffusions are universal algorithms for private optimization: they provide optimal results for DP-ERM and DP-SCO, under pure and approximate DP, for both convex and strongly convex losses.

Auditing the stability of a machine learning model to small changes in the training procedure is critical for engendering trust in practical applications. For example, a model should not be overly sensitive to removing a small fraction of its training data. However, algorithmically validating this property seems computationally challenging, even for the simplest of models: Ordinary Least Squares (OLS) linear regression. Concretely, recent work defines the stability of a regression as the minimum number of samples that need to be removed so that rerunning the analysis overturns the conclusion (Broderick et al., 2020), specifically meaning that the sign of a particular coefficient of the OLS regressor changes. But the only known approach for estimating this metric, besides the obvious exponential-time algorithm, is a greedy heuristic that may produce severe overestimates and therefore cannot certify stability. We show that stability can be efficiently certified in the low-dimensional regime: when the number of covariates is a constant but the number of samples is large, there are polynomial-time algorithms for estimating (a fractional version of) stability, with provable approximation guarantees. Applying our algorithms to the Boston Housing dataset, we exhibit regression analyses where our estimator outperforms the greedy heuristic, and can successfully certify stability even in the regime where a constant fraction of the samples are dropped.

We develop provable and efficient algorithms for estimating stability of OLS to dropping samples in the low-dimensional regime.