ICML 2024 论文列表

We consider the problem of quantifying the amount of influence one agent can exert on another in the setting of multi-agent reinforcement learning (MARL). As a step towards a unified approach to express agents' interdependencies, we introduce the total and state influence measurement functions. Both of these are valid for all common MARL systems, such as the discounted reward setting. Additionally, we propose novel quantities, called the total impact measurement (TIM) and state impact measurement (SIM), that characterize one agent's influence on another by the maximum impact it can have on the other agents' expected returns and represent instances of impact measurement functions in the average reward setting. Furthermore, we provide approximation algorithms for TIM and SIM with simultaneously learning approximations of agents' expected returns, error bounds, stability analyses under changes of the policies, and convergence guarantees. The approximation algorithm relies only on observing other agents' actions and is, other than that, fully decentralized. Through empirical studies, we validate our approach's effectiveness in identifying intricate influence structures in complex interactions. Our work appears to be the first study of determining influence structures in the multi-agent average reward setting with convergence guarantees.

A key challenge in Meta-Reinforcement Learning (meta-RL) is the task distribution shift, since the generalization ability of most current meta-RL methods is limited to tasks sampled from the training distribution. In this paper, we propose Posterior Sampling Bayesian Lifelong In-Context Reinforcement Learning (PSBL), which is robust to task distribution shift. PSBL meta-trains a variant of transformer to directly perform amortized inference about the Predictive Posterior Distribution (PPD) of the optimal policy. Once trained, the network can infer the PPD online with frozen parameters. The agent then samples actions from the approximate PPD to perform online exploration, which progressively reduces uncertainty and enhances performance in the interaction with the environment. This property is known as in-context learning. Experimental results demonstrate that PSBL significantly outperforms standard Meta RL methods both in tasks with sparse rewards and dense rewards when the test task distribution is strictly shifted from the training distribution.

We focus on constrained, $L$-smooth, potentially stochastic and nonconvex-nonconcave min-max problems either satisfying $\rho$-cohypomonotonicity or admitting a solution to the $\rho$-weakly Minty Variational Inequality (MVI), where larger values of the parameter $\rho>0$ correspond to a greater degree of nonconvexity. These problem classes include examples in two player reinforcement learning, interaction dominant min-max problems, and certain synthetic test problems on which classical min-max algorithms fail. It has been conjectured that first-order methods can tolerate a value of $\rho$ no larger than $\frac{1}{L}$, but existing results in the literature have stagnated at the tighter requirement $\rho < \frac{1}{2L}$. With a simple argument, we obtain optimal or best-known complexity guarantees with cohypomonotonicity or weak MVI conditions for $\rho < \frac{1}{L}$. First main insight for the improvements in the convergence analyses is to harness the recently proposed *conic nonexpansiveness* property of operators. Second, we provide a refined analysis for inexact Halpern iteration that relaxes the required inexactness level to improve some state-of-the-art complexity results even for constrained stochastic convex-concave min-max problems. Third, we analyze a stochastic inexact Krasnosel'skii-Mann iteration with a multilevel Monte Carlo estimator when the assumptions only hold with respect to a solution.

We study inference on the long-term causal effect of a continual exposure to a novel intervention, which we term a long-term treatment, based on an experiment involving only short-term observations. Key examples include the long-term health effects of regularly-taken medicine or of environmental hazards and the long-term effects on users of changes to an online platform. This stands in contrast to short-term treatments or "shocks," whose long-term effect can reasonably be mediated by short-term observations, enabling the use of surrogate methods. Long-term treatments by definition have direct effects on long-term outcomes via continual exposure, so surrogacy conditions cannot reasonably hold. We connect the problem with offline reinforcement learning, leveraging doubly-robust estimators to estimate long-term causal effects for long-term treatments and construct confidence intervals.

In this work, we introduce PIPER: Primitive-Informed Preference-based Hierarchical reinforcement learning via Hindsight Relabeling, a novel approach that leverages preference-based learning to learn a reward model, and subsequently uses this reward model to relabel higher-level replay buffers. Since this reward is unaffected by lower primitive behavior, our relabeling-based approach is able to mitigate non-stationarity, which is common in existing hierarchical approaches, and demonstrates impressive performance across a range of challenging sparse-reward tasks. Since obtaining human feedback is typically impractical, we propose to replace the human-in-the-loop approach with our primitive-in-the-loop approach, which generates feedback using sparse rewards provided by the environment. Moreover, in order to prevent infeasible subgoal prediction and avoid degenerate solutions, we propose primitive-informed regularization that conditions higher-level policies to generate feasible subgoals for lower-level policies. We perform extensive experiments to show that PIPER mitigates non-stationarity in hierarchical reinforcement learning and achieves greater than 50$\\%$ success rates in challenging, sparse-reward robotic environments, where most other baselines fail to achieve any significant progress.

In this paper, we derive an upper bound of the difference between a kernel matrix and its expectation under a mild assumption. Specifically, we assume that the true distribution of the training data is an unknown isotropic Gaussian distribution. When the kernel function is a Gaussian kernel, and the mean of each cluster is sufficiently separated, we find that the expectation of a kernel matrix can be close to a rank-$k$ matrix, where $k$ is the cluster number. Moreover, we prove that the normalized kernel matrix of the training set deviates (w.r.t. Frobenius norm) from its expectation in the order of $\widetilde{\mathcal{O}}(1/\sqrt{d})$, where $d$ is the dimension of samples. Based on the above theoretical results, we propose a novel multiple kernel clustering framework which attempts to learn the information of the expectation kernel matrices. First, we aim to minimize the distance between each base kernel and a rank-$k$ matrix, which is a proxy of the expectation kernel. Then, we fuse these rank-$k$ matrices into a consensus rank-$k$ matrix to find the clustering structure. Using an anchor-based method, the proposed framework is flexible with the sizes of input kernel matrices and able to handle large-scale datasets. We also provide the approximation guarantee by deriving two non-asymptotic bounds for the consensus kernel and clustering indicator matrices. Finally, we conduct extensive experiments to verify the clustering performance of the proposed method and the correctness of the proposed theoretical results.

The goal-reaching tasks with safety constraints are common control problems in real world, such as intelligent driving and robot manipulation. The difficulty of this kind of problem comes from the exploration termination caused by safety constraints and the sparse rewards caused by goals. The existing safe RL avoids unsafe exploration by restricting the search space to a feasible region, the essence of which is the pruning of the search space. However, there are still many ineffective explorations in the feasible region because of the ignorance of the goals. Our approach considers both safety and goals; the policy space pruning is achieved by a function called feasible reachable function, which describes whether there is a policy to make the agent safely reach the goals in the finite time domain. This function naturally satisfies the self-consistent condition and the risky Bellman equation, which can be solved by the fixed point iteration method. On this basis, we propose feasible reachable policy iteration (FRPI), which is divided into three steps: policy evaluation, region expansion, and policy improvement. In the region expansion step, by using the information of agent to reach the goals, the convergence of the feasible region is accelerated, and simultaneously a smaller feasible reachable region is identified. The experimental results verify the effectiveness of the proposed FR function in both improving the convergence speed of better or comparable performance without sacrificing safety and identifying a smaller policy space with higher sample efficiency.

Most reinforcement learning practitioners evaluate their policies with online Monte Carlo estimators for either hyperparameter tuning or testing different algorithmic design choices, where the policy is repeatedly executed in the environment to get the average outcome. Such massive interactions with the environment are prohibitive in many scenarios. In this paper, we propose novel methods that improve the data efficiency of online Monte Carlo estimators while maintaining their unbiasedness. We first propose a tailored closed-form behavior policy that provably reduces the variance of an online Monte Carlo estimator. We then design efficient algorithms to learn this closed-form behavior policy from previously collected offline data. Theoretical analysis is provided to characterize how the behavior policy learning error affects the amount of reduced variance. Compared with previous works, our method achieves better empirical performance in a broader set of environments, with fewer requirements for offline data.

In this paper, we prove that Distributional Reinforcement Learning (DistRL), which learns the return distribution, can obtain second-order bounds in both online and offline RL in general settings with function approximation. Second-order bounds are instance-dependent bounds that scale with the variance of return, which we prove are tighter than the previously known small-loss bounds of distributional RL. To the best of our knowledge, our results are the first second-order bounds for low-rank MDPs and for offline RL. When specializing to contextual bandits (one-step RL problem), we show that a distributional learning based optimism algorithm achieves a second-order worst-case regret bound, and a second-order gap dependent bound, simultaneously. We also empirically demonstrate the benefit of DistRL in contextual bandits on real-world datasets. We highlight that our analysis with DistRL is relatively simple, follows the general framework of optimism in the face of uncertainty and does not require weighted regression. Our results suggest that DistRL is a promising framework for obtaining second-order bounds in general RL settings, thus further reinforcing the benefits of DistRL.

Estimating the parameters of a probabilistic directed graphical model from incomplete data is a long-standing challenge. This is because, in the presence of latent variables, both the likelihood function and posterior distribution are intractable without assumptions about structural dependencies or model classes. While existing learning methods are fundamentally based on likelihood maximization, here we offer a new view of the parameter learning problem through the lens of optimal transport. This perspective licenses a general framework that operates on any directed graphs without making unrealistic assumptions on the posterior over the latent variables or resorting to variational approximations. We develop a theoretical framework and support it with extensive empirical evidence demonstrating the versatility and robustness of our approach. Across experiments, we show that not only can our method effectively recover the ground-truth parameters but it also performs comparably or better than competing baselines on downstream applications.

Training autonomous agents with sparse rewards is a long-standing problem in online reinforcement learning (RL), due to low data efficiency. Prior work overcomes this challenge by extracting useful knowledge from offline data, often accomplished through the learning of action distribution from offline data and utilizing the learned distribution to facilitate online RL. However, since the offline data are given and fixed, the extracted knowledge is inherently limited, making it difficult to generalize to new tasks. We propose a novel approach that leverages offline data to learn a generative diffusion model, coined as Adaptive Trajectory Diffuser (ATraDiff). This model generates synthetic trajectories, serving as a form of data augmentation and consequently enhancing the performance of online RL methods. The key strength of our diffuser lies in its adaptability, allowing it to effectively handle varying trajectory lengths and mitigate distribution shifts between online and offline data. Because of its simplicity, ATraDiff seamlessly integrates with a wide spectrum of RL methods. Empirical evaluation shows that ATraDiff consistently achieves state-of-the-art performance across a variety of environments, with particularly pronounced improvements in complicated settings. Our code and demo video are available at https://atradiff.github.io.

In-context learning is a promising approach for offline reinforcement learning (RL) to handle online tasks, which can be achieved by providing task prompts. Recent works demonstrated that in-context RL could emerge with self-improvement in a trial-and-error manner when treating RL tasks as an across-episodic sequential prediction problem. Despite the self-improvement not requiring gradient updates, current works still suffer from high computational costs when the across-episodic sequence increases with task horizons. To this end, we propose an In-context Decision Transformer (IDT) to achieve self-improvement in a high-level trial-and-error manner. Specifically, IDT is inspired by the efficient hierarchical structure of human decision-making and thus reconstructs the sequence to consist of high-level decisions instead of low-level actions that interact with environments. As one high-level decision can guide multi-step low-level actions, IDT naturally avoids excessively long sequences and solves online tasks more efficiently. Experimental results show that IDT achieves state-of-the-art in long-horizon tasks over current in-context RL methods. In particular, the online evaluation time of our IDT is 36$\times$ times faster than baselines in the D4RL benchmark and 27$\times$ times faster in the Grid World benchmark.

Offline reinforcement learning algorithms have proven effective on datasets highly connected to the target downstream task. Yet, by leveraging a novel testbed (MOOD) in which trajectories come from heterogeneous sources, we show that existing methods struggle with diverse data: their performance considerably deteriorates as data collected for related but different tasks is simply added to the offline buffer. In light of this finding, we conduct a large empirical study where we formulate and test several hypotheses to explain this failure. Surprisingly, we find that targeted scale, more than algorithmic considerations, is the key factor influencing performance. We show that simple methods like AWAC and IQL with increased policy size overcome the paradoxical failure modes from the inclusion of additional data in MOOD, and notably outperform prior state-of-the-art algorithms on the canonical D4RL benchmark.

Active learning is increasingly adopted for expensive multi-objective combinatorial optimization problems, but it involves a challenging subset selection problem, optimizing the batch acquisition score that quantifies the goodness of a batch for evaluation. Due to the excessively large search space of the subset selection problem, prior methods optimize the batch acquisition on the latent space, which has discrepancies with the actual space, or optimize individual acquisition scores without considering the dependencies among candidates in a batch instead of directly optimizing the batch acquisition. To manage the vast search space, a simple and effective approach is the greedy method, which decomposes the problem into smaller subproblems, yet it has difficulty in parallelization since each subproblem depends on the outcome from the previous ones. To this end, we introduce a novel greedy-style subset selection algorithm that optimizes batch acquisition directly on the combinatorial space by sequential greedy sampling from the greedy policy, specifically trained to address all greedy subproblems concurrently. Notably, our experiments on the red fluorescent proteins design task show that our proposed method achieves the baseline performance in 1.69x fewer queries, demonstrating its efficiency.

We focus on the task of learning the value function in the reinforcement learning (RL) setting. This task is often solved by updating a pair of online and target networks while ensuring that the parameters of these two networks are equivalent. We propose Lookahead-Replicate (LR), a new value-function approximation algorithm that is agnostic to this parameter-space equivalence. Instead, the LR algorithm is designed to maintain an equivalence between the two networks in the function space. This value-based equivalence is obtained by employing a new target-network update. We show that LR leads to a convergent behavior in learning the value function. We also present empirical results demonstrating that LR-based target-network updates significantly improve deep RL on the Atari benchmark.

Multistep returns, such as $n$-step returns and $\lambda$-returns, are commonly used to improve the sample efficiency of reinforcement learning (RL) methods. The variance of the multistep returns becomes the limiting factor in their length; looking too far into the future increases variance and reverses the benefits of multistep learning. In our work, we demonstrate the ability of compound returns—weighted averages of $n$-step returns—to reduce variance. We prove for the first time that any compound return with the same contraction modulus as a given $n$-step return has strictly lower variance. We additionally prove that this variance-reduction property improves the finite-sample complexity of temporal-difference learning under linear function approximation. Because general compound returns can be expensive to implement, we introduce two-bootstrap returns which reduce variance while remaining efficient, even when using minibatched experience replay. We conduct experiments showing that compound returns often increase the sample efficiency of $n$-step deep RL agents like DQN and PPO.

Recent studies on online reinforcement learning (RL) have demonstrated the advantages of learning multiple behaviors from a single task, as in the case of few-shot adaptation to a new environment. Although this approach is expected to yield similar benefits in offline RL, appropriate methods for learning multiple solutions have not been fully investigated in previous studies. In this study, we therefore addressed the problem of finding multiple solutions from a single task in offline RL. We propose algorithms that can learn multiple solutions in offline RL, and empirically investigate their performance. Our experimental results show that the proposed algorithm learns multiple qualitatively and quantitatively distinctive solutions in offline RL.

We propose a novel algorithmic framework for distributional reinforcement learning, based on learning finite-dimensional mean embeddings of return distributions. The framework reveals a wide variety of new algorithms for dynamic programming and temporal-difference algorithms that rely on the sketch Bellman operator, which updates mean embeddings with simple linear-algebraic computations. We provide asymptotic convergence theory, and examine the empirical performance of the algorithms on a suite of tabular tasks. Further, we show that this approach can be straightforwardly combined with deep reinforcement learning.

In recent years, Reinforcement Learning (RL) has shown great promise in session-based recommendation. Sequential models that use RL have reached state-of-the-art performance for the Next-item Prediction (NIP) task. This result is intriguing, as the NIP task only evaluates how well the system can correctly recommend the next item to the user, while the goal of RL is to find a policy that optimizes rewards in the long term -- sometimes at the expense of suboptimal short-term performance. Then, how can RL improve the system's performance on short-term metrics? This article investigates this question by exploring proxy learning objectives, which we identify as goals RL models might be following, and thus could explain the performance boost. We found that RL -- when used as an auxiliary loss -- promotes the learning of embeddings that capture information about the user's previously interacted items. Subsequently, we replaced the RL objective with a straightforward auxiliary loss designed to predict the number of items the user interacted with. This substitution results in performance gains comparable to RL. These findings pave the way to improve performance and understanding of RL methods for recommender systems.