ICML 2024 Papers

Learning a high-quality transition model is of great importance for sequential decision-making tasks, especially in offline settings. Nevertheless, the complex behaviors of transition dynamics in real-world environments pose challenges for the standard forward models because of their inductive bias towards smooth regressors, conflicting with the inherent nature of transitions such as discontinuity or large curvature. In this work, we propose to model the transition probability implicitly through a scalar-value energy function, which enables not only flexible distribution prediction but also capturing complex transition behaviors. The Energy-based Transition Models (ETM) are shown to accurately fit the discontinuous transition functions and better generalize to out-of-distribution transition data. Furthermore, we demonstrate that energy-based transition models improve the evaluation accuracy and significantly outperform other off-policy evaluation methods in DOPE benchmark. Finally, we show that energy-based transition models also benefit reinforcement learning and outperform prior offline RL algorithms in D4RL Gym-Mujoco tasks.

Temporal difference (TD) learning algorithms with neural network function parameterization have well-established empirical success in many practical large-scale reinforcement learning tasks. However, theoretical understanding of these algorithms remains challenging due to the nonlinearity of the action-value approximation. In this paper, we develop an improved non-asymptotic analysis of the neural TD method with a general $L$-layer neural network. New proof techniques are developed and an improved new $\tilde{\mathcal{O}}(\epsilon^{-1})$ sample complexity is derived. To our best knowledge, this is the first finite-time analysis of neural TD that achieves an $\tilde{\mathcal{O}}(\epsilon^{-1})$ complexity under the Markovian sampling, as opposed to the best known $\tilde{\mathcal{O}}(\epsilon^{-2})$ complexity in the existing literature.

Partial monitoring is a generic framework of online decision-making problems with limited feedback. To make decisions from such limited feedback, it is necessary to find an appropriate distribution for exploration. Recently, a powerful approach for this purpose, exploration by optimization (ExO), was proposed, which achieves optimal bounds in adversarial environments with follow-the-regularized-leader for a wide range of online decision-making problems. However, a naive application of ExO in stochastic environments significantly degrades regret bounds. To resolve this issue in locally observable games, we first establish a new framework and analysis for ExO with a hybrid regularizer. This development allows us to significantly improve existing regret bounds of best-of-both-worlds (BOBW) algorithms, which achieves nearly optimal bounds both in stochastic and adversarial environments. In particular, we derive a stochastic regret bound of $O(\sum_{a \neq a^*} k^2 m^2 \log T / \Delta_a)$, where $k$, $m$, and $T$ are the numbers of actions, observations and rounds, $a^*$ is an optimal action, and $\Delta_a$ is the suboptimality gap for action $a$. This bound is roughly $\Theta(k^2 \log T)$ times smaller than existing BOBW bounds. In addition, for globally observable games, we provide a new BOBW algorithm with the first $O(\log T)$ stochastic bound.

Reinforcement Learning (RL) for constrained MDPs (CMDPs) is an increasingly important problem for various applications. Often, the average criterion is more suitable than the discounted criterion. Yet, RL for average-CMDPs (ACMDPs) remains a challenging problem. Algorithms designed for discounted constrained RL problems often do not perform well for the average CMDP setting. In this paper, we introduce a new policy optimization with function approximation algorithm for constrained MDPs with the average criterion. The Average-Constrained Policy Optimization (ACPO) algorithm is inspired by trust region-based policy optimization algorithms. We develop basic sensitivity theory for average CMDPs, and then use the corresponding bounds in the design of the algorithm. We provide theoretical guarantees on its performance, and through extensive experimental work in various challenging OpenAI Gym environments, show its superior empirical performance when compared to other state-of-the-art algorithms adapted for the ACMDPs.

We have developed a new framework based on the theory of causal inference to protect language models against backdoor attacks. Backdoor attackers can poison language models with different types of triggers, such as words, sentences, grammar, and style, enabling them to selectively modify the decision-making of the victim model. However, existing defense approaches are only effective when the backdoor attack form meets specific assumptions, making it difficult to counter diverse backdoor attacks. We propose a new defense framework **F**ront-door **A**djustment for **B**ackdoor **E**limination (FABE) based on causal reasoning that does not rely on assumptions about the form of triggers. This method effectively differentiates between spurious and legitimate associations by creating a 'front door' that maps out the actual causal relationships. The term 'front door' refers to a text that retains the semantic equivalence of the initial input, which is generated by an additional, fine-tuned language model, denoted as the defense model. Our defense experiments against various attack methods at the token, sentence, and syntactic levels reduced the attack success rate from 93.63% to 15.12%, improving the defense effect by 2.91 times compared to the best baseline result of 66.61%, achieving state-of-the-art results. Through ablation study analysis, we analyzed the effect of each module in FABE, demonstrating the importance of complying with the front-door criterion and front-door adjustment formula, which also explains why previous methods failed. Our code to reproduce the experiments is available at: https://github.com/lyr17/Frontdoor-Adjustment-Backdoor-Elimination.

The resonate-and-fire (RF) neuron, introduced over two decades ago, is a simple, efficient, yet biologically plausible spiking neuron model, which can extract frequency patterns within the time domain due to its resonating membrane dynamics. However, previous RF formulations suffer from intrinsic shortcomings that limit effective learning and prevent exploiting the principled advantage of RF neurons. Here, we introduce the balanced RF (BRF) neuron, which alleviates some of the intrinsic limitations of vanilla RF neurons and demonstrates its effectiveness within recurrent spiking neural networks (RSNNs) on various sequence learning tasks. We show that networks of BRF neurons achieve overall higher task performance, produce only a fraction of the spikes, and require significantly fewer parameters as compared to modern RSNNs. Moreover, BRF-RSNN consistently provide much faster and more stable training convergence, even when bridging many hundreds of time steps during backpropagation through time (BPTT). These results underscore that our BRF-RSNN is a strong candidate for future large-scale RSNN architectures, further lines of research in SNN methodology, and more efficient hardware implementations.

Despite the recent successes of multi-agent reinforcement learning (MARL) algorithms, efficiently adapting to co-players in mixed-motive environments remains a significant challenge. One feasible approach is to hierarchically model co-players' behavior based on inferring their characteristics. However, these methods often encounter difficulties in efficient reasoning and utilization of inferred information. To address these issues, we propose Hierarchical Opponent modeling and Planning (HOP), a novel multi-agent decision-making algorithm that enables few-shot adaptation to unseen policies in mixed-motive environments. HOP is hierarchically composed of two modules: an opponent modeling module that infers others' goals and learns corresponding goal-conditioned policies, and a planning module that employs Monte Carlo Tree Search (MCTS) to identify the best response. Our approach improves efficiency by updating beliefs about others' goals both across and within episodes and by using information from the opponent modeling module to guide planning. Experimental results demonstrate that in mixed-motive environments, HOP exhibits superior few-shot adaptation capabilities when interacting with various unseen agents, and excels in self-play scenarios. Furthermore, the emergence of social intelligence during our experiments underscores the potential of our approach in complex multi-agent environments.

Real-world datasets often exhibit long-tailed distributions, compromising the generalization and fairness of learning-based models. This issue is particularly pronounced in Image Aesthetics Assessment (IAA) tasks, where such imbalance is difficult to mitigate due to a severe distribution mismatch between features and labels, as well as the great sensitivity of aesthetics to image variations. To address these issues, we propose an Enhancer against Long-Tail for Aesthetics-oriented models (ELTA). ELTA first utilizes a dedicated mixup technique to enhance minority feature representation in high-level space while preserving their intrinsic aesthetic qualities. Next, it aligns features and labels through a similarity consistency approach, effectively alleviating the distribution mismatch. Finally, ELTA adopts a specific strategy to refine the output distribution, thereby enhancing the quality of pseudo-labels. Experiments on four representative datasets (AVA, AADB, TAD66K, and PARA) show that our proposed ELTA achieves state-of-the-art performance by effectively mitigating the long-tailed issue in IAA datasets. Moreover, ELTA is designed with plug-and-play capabilities for seamless integration with existing methods. To our knowledge, this is the first contribution in the IAA community addressing long-tail. All resources are available in here.

Post-training quantization of Large Language Models (LLMs) is challenging. In this work, we introduce **L**ow-rank **Q**uantization **E**rror **R**eduction (LQER), which combines quantization and low-rank approximation to recover the model capability. LQER leverages an activation-induced scale matrix to drive the singular value distribution of quantization error towards a desirable distribution, which enables nearly-lossless W4A8 quantization on various LLMs and downstream tasks without the need for knowledge distillation, grid search, or gradient-based iterative optimization. Unlike existing methods, the computation pattern of LQER eliminates the need for specialized Scatter and Gather processes to collect high-precision weights from irregular memory locations. Our W4A8 LLMs achieve near-lossless performance on six popular downstream tasks, while using $1.36 \times$ fewer hardware resources than the leading state-of-the-art method. We will open-source our framework at [https://github.com/ChengZhang-98/lqer](https://github.com/ChengZhang-98/lqer)

We study the problem of forecasting the evolution of a function of the state (observable) of a discrete ergodic dynamical system over multiple time steps. The elegant theory of Koopman and transfer operators can be used to evolve any such function forward in time. However, their estimators are usually unreliable in long-term forecasting. We show how classical techniques of eigenvalue deflation from operator theory and feature centering from statistics can be exploited to enhance standard estimators. We develop a novel technique to derive high probability bounds on powers of empirical estimators. Our approach, rooted in the stability _theory of non-normal operators_, allows us to establish uniform in time bounds for the forecasting error, which hold even on _infinite time horizons_. We further show that our approach can be seamlessly employed to forecast future state distributions from an initial one, with provably uniform error bounds. Numerical experiments illustrate the advantages of our approach in practice.

Deep equilibrium models (DEQs), as typical implicit neural networks, have demonstrated remarkable success on various tasks. There is, however, a lack of theoretical understanding of the connections and differences between implicit DEQs and explicit neural network models. In this paper, leveraging recent advances in random matrix theory (RMT), we perform an in-depth analysis on the eigenspectra of the conjugate kernel (CK) and neural tangent kernel (NTK) matrices for implicit DEQs, when the input data are drawn from a high-dimensional Gaussia mixture. We prove that, in this setting, the spectral behavior of these Implicit-CKs and NTKs depend on the DEQ activation function and initial weight variances, *but only via a system of four nonlinear equations*. As a direct consequence of this theoretical result, we demonstrate that a shallow explicit network can be carefully designed to produce the same CK or NTK as a given DEQ. Despite derived here for Gaussian mixture data, empirical results show the proposed theory and design principles also apply to popular real-world datasets.

The label smoothness assumption is at the core of Graph Convolutional Networks (GCNs): nodes in a local region have similar labels. Thus, GCN performs local feature smoothing operation to adhere to this assumption. However, there exist some nodes whose labels obtained by feature smoothing conflict with the label smoothness assumption. We find that the label smoothness assumption and the process of feature smoothing are both problematic on these nodes, and call these nodes out of GCN's control (OOC nodes). In this paper, first, we design the corresponding algorithm to locate the OOC nodes, then we summarize the characteristics of OOC nodes that affect their representation learning, and based on their characteristics, we present DaGCN, an efficient framework that can facilitate the OOC nodes. Extensive experiments verify the superiority of the proposed method and demonstrate that current advanced GCNs are improvements specifically on OOC nodes; the remaining nodes under GCN's control (UC nodes) are already optimally represented by vanilla GCN on most datasets.

Modeling the complex three-dimensional (3D) dynamics of relational systems is an important problem in the natural sciences, with applications ranging from molecular simulations to particle mechanics. Machine learning methods have achieved good success by learning graph neural networks to model spatial interactions. However, these approaches do not faithfully capture temporal correlations since they only model next-step predictions. In this work, we propose Equivariant Graph Neural Operator (EGNO), a novel and principled method that directly models dynamics as trajectories instead of just next-step prediction. Different from existing methods, EGNO explicitly learns the temporal evolution of 3D dynamics where we formulate the dynamics as a function over time and learn neural operators to approximate it. To capture the temporal correlations while keeping the intrinsic SE(3)-equivariance, we develop equivariant temporal convolutions parameterized in the Fourier space and build EGNO by stacking the Fourier layers over equivariant networks. EGNO is the first operator learning framework that is capable of modeling solution dynamics functions over time while retaining 3D equivariance. Comprehensive experiments in multiple domains, including particle simulations, human motion capture, and molecular dynamics, demonstrate the significantly superior performance of EGNO against existing methods, thanks to the equivariant temporal modeling. Our code is available at https://github.com/MinkaiXu/egno.

Multimodal Large Language Models (MLLMs) have gained significant attention recently, showing remarkable potential in artificial general intelligence. However, assessing the utility of MLLMs presents considerable challenges, primarily due to the absence multimodal benchmarks that align with human preferences. Drawing inspiration from the concept of LLM-as-a-Judge within LLMs, this paper introduces a novel benchmark, termed MLLM-as-a-Judge, to assess the ability of MLLMs in assisting judges across diverse modalities, encompassing three distinct tasks: Scoring Evaluation, Pair Comparison, and Batch Ranking. Our study reveals that, while MLLMs demonstrate remarkable human-like discernment in Pair Comparisons, there is a significant divergence from human preferences in Scoring Evaluation and Batch Ranking tasks. Furthermore, a closer examination reveals persistent challenges in the evaluative capacities of LLMs, including diverse biases, hallucinatory responses, and inconsistencies in judgment, even in advanced models such as GPT-4V. These findings emphasize the pressing need for enhancements and further research efforts to be undertaken before regarding MLLMs as fully reliable evaluators. In light of this, we advocate for additional efforts dedicated to supporting the continuous development within the domain of MLLM functioning as judges. The code and dataset are publicly available at our project homepage: https://mllm-judge.github.io/.

Mean-field variational inference (MFVI) methods provide computationally cheap approximations to the posterior of Bayesian Neural Networks (BNNs) when compared to alternatives like MCMC. However, applying MFVI to BNNs encounters limitations due to the Monte Carlo sampling problem. This problem stems from two main issues. *First*, most samples do not accurately represent the most probable weights. *Second*, random sampling from variational distributions introduces high variance in gradient estimates, which can hinder the optimization process, leading to slow convergence or even failure. In this paper, we introduce a novel sampling method called *Reparameterized Importance Sampling* (RIS) to estimate the first moment in neural networks, reducing variance during feed-forward propagation. We begin by analyzing the generalized form of the optimal proposal distribution and presenting an inexpensive approximation. Next, we describe the sampling process from the proposal distribution as a transformation that combines exogenous randomness with the variational parameters. Our experimental results demonstrate the effectiveness of the proposed RIS method in three critical aspects: improved convergence, enhanced predictive performance, and successful uncertainty estimation for out-of-distribution data.

Pre-trained language models (LMs) are able to perform complex reasoning without explicit fine-tuning. To understand how pre-training with a next-token prediction objective contributes to the emergence of such reasoning capability, we propose that we can view an LM as deriving new conclusions by aggregating indirect reasoning paths seen at pre-training time. We found this perspective effective in two important cases of reasoning: logic reasoning with knowledge graphs (KGs) and chain-of-thought (CoT) reasoning. More specifically, we formalize the reasoning paths as random walk paths on the knowledge/reasoning graphs. Analyses of learned LM distributions suggest that a weighted sum of relevant random walk path probabilities is a reasonable way to explain how LMs reason. Experiments and analysis on multiple KG and CoT datasets reveal the effect of training on random walk paths and suggest that augmenting unlabeled random walk reasoning paths can improve real-world multi-step reasoning performance.

We study the statistical complexity of offline decision-making with function approximation, establishing (near) minimax-optimal rates for stochastic contextual bandits and Markov decision processes. The performance limits are captured by the pseudo-dimension of the (value) function class and a new characterization of the behavior policy that *strictly* subsumes all the previous notions of data coverage in the offline decision-making literature. In addition, we seek to understand the benefits of using offline data in online decision-making and show nearly minimax-optimal rates in a wide range of regimes.

Many processes in biology and drug discovery involve various 3D interactions between molecules, such as protein and protein, protein and small molecule, etc. Given that different molecules are usually represented in different granularity, existing methods usually encode each type of molecules independently with different models, leaving it defective to learn the various underlying interaction physics. In this paper, we first propose to universally represent an arbitrary 3D complex as a geometric graph of sets, shedding light on encoding all types of molecules with one model. We then propose a Generalist Equivariant Transformer (GET) to effectively capture both domain-specific hierarchies and domain-agnostic interaction physics. To be specific, GET consists of a bilevel attention module, a feed-forward module and a layer normalization module, where each module is E(3) equivariant and specialized for handling sets of variable sizes. Notably, in contrast to conventional pooling-based hierarchical models, our GET is able to retain fine-grained information of all levels. Extensive experiments on the interactions between proteins, small molecules and RNA/DNAs verify the effectiveness and generalization capability of our proposed method across different domains.

In this paper, we explore a new generative approach for learning visual representations. Our method, DARL, employs a decoder-only Transformer to predict image patches autoregressively. We find that training with Mean Squared Error (MSE) alone leads to strong representations. To enhance the image generation ability, we replace the MSE loss with the diffusion objective by using a denoising patch decoder. We show that the learned representation can be improved by using tailored noise schedules and longer training in larger models. Notably, the optimal schedule differs significantly from the typical ones used in standard image diffusion models. Overall, despite its simple architecture, DARL delivers performance remarkably close to state-of-the-art masked prediction models under the fine-tuning protocol. This marks an important step towards a unified model capable of both visual perception and generation, effectively combining the strengths of autoregressive and denoising diffusion models.