Zongzhang Zhang

Guoqing Liu, Di Xue, Shufang Xie et al.

Retrosynthesis, which aims to find a route to synthesize a target molecule from commercially available starting materials, is a critical task in drug discovery and materials design. Recently, the combination of ML-based single-step reaction predictors with multi-step planners has led to promising results. However, the single-step predictors are mostly trained offline to optimize the single-step accuracy, without considering complete routes. Here, we leverage reinforcement learning (RL) to improve the single-step predictor, by using a tree-shaped MDP to optimize complete routes. Specifically, we propose a novel online training algorithm, called Planning with Dual Value Networks (PDVN), which alternates between the planning phase and updating phase. In PDVN, we construct two separate value networks to predict the synthesizability and cost of molecules, respectively. To maintain the single-step accuracy, we design a two-branch network structure for the single-step predictor. On the widely-used USPTO dataset, our PDVN algorithm improves the search success rate of existing multi-step planners (e.g., increasing the success rate from 85.79% to 98.95% for Retro*, and reducing the number of model calls by half while solving 99.47% molecules for RetroGraph). Additionally, PDVN helps find shorter synthesis routes (e.g., reducing the average route length from 5.76 to 4.83 for Retro*, and from 5.63 to 4.78 for RetroGraph). Our code is available at \url{https://github.com/DiXue98/PDVN}.

Rongjun Qin, Feng Chen, Tonghan Wang et al. · harvard, tsinghua

Team adaptation to new cooperative tasks is a hallmark of human intelligence, which has yet to be fully realized in learning agents. Previous work on multi-agent transfer learning accommodate teams of different sizes, heavily relying on the generalization ability of neural networks for adapting to unseen tasks. We believe that the relationship among tasks provides the key information for policy adaptation. In this paper, we try to discover and exploit common structures among tasks for more efficient transfer, and propose to learn effect-based task representations as a common space of tasks, using an alternatively fixed training scheme. We demonstrate that the task representation can capture the relationship among tasks, and can generalize to unseen tasks. As a result, the proposed method can help transfer learned cooperation knowledge to new tasks after training on a few source tasks. We also find that fine-tuning the transferred policies help solve tasks that are hard to learn from scratch.

Yuhang Ran, Yi-Chen Li, Fuxiang Zhang et al.

We consider the problem of learning the best possible policy from a fixed dataset, known as offline Reinforcement Learning (RL). A common taxonomy of existing offline RL works is policy regularization, which typically constrains the learned policy by distribution or support of the behavior policy. However, distribution and support constraints are overly conservative since they both force the policy to choose similar actions as the behavior policy when considering particular states. It will limit the learned policy's performance, especially when the behavior policy is sub-optimal. In this paper, we find that regularizing the policy towards the nearest state-action pair can be more effective and thus propose Policy Regularization with Dataset Constraint (PRDC). When updating the policy in a given state, PRDC searches the entire dataset for the nearest state-action sample and then restricts the policy with the action of this sample. Unlike previous works, PRDC can guide the policy with proper behaviors from the dataset, allowing it to choose actions that do not appear in the dataset along with the given state. It is a softer constraint but still keeps enough conservatism from out-of-distribution actions. Empirical evidence and theoretical analysis show that PRDC can alleviate offline RL's fundamentally challenging value overestimation issue with a bounded performance gap. Moreover, on a set of locomotion and navigation tasks, PRDC achieves state-of-the-art performance compared with existing methods. Code is available at https://github.com/LAMDA-RL/PRDC

Chen-Xiao Gao, Chenyang Wu, Mingjun Cao et al.

Decision Transformer (DT), which employs expressive sequence modeling techniques to perform action generation, has emerged as a promising approach to offline policy optimization. However, DT generates actions conditioned on a desired future return, which is known to bear some weaknesses such as the susceptibility to environmental stochasticity. To overcome DT's weaknesses, we propose to empower DT with dynamic programming. Our method comprises three steps. First, we employ in-sample value iteration to obtain approximated value functions, which involves dynamic programming over the MDP structure. Second, we evaluate action quality in context with estimated advantages. We introduce two types of advantage estimators, IAE and GAE, which are suitable for different tasks. Third, we train an Advantage-Conditioned Transformer (ACT) to generate actions conditioned on the estimated advantages. Finally, during testing, ACT generates actions conditioned on a desired advantage. Our evaluation results validate that, by leveraging the power of dynamic programming, ACT demonstrates effective trajectory stitching and robust action generation in spite of the environmental stochasticity, outperforming baseline methods across various benchmarks. Additionally, we conduct an in-depth analysis of ACT's various design choices through ablation studies. Our code is available at https://github.com/LAMDA-RL/ACT.

Xu-Hui Liu, Feng Xu, Xinyu Zhang et al.

Imitation learning aims to mimic the behavior of experts without explicit reward signals. Passive imitation learning methods which use static expert datasets typically suffer from compounding error, low sample efficiency, and high hyper-parameter sensitivity. In contrast, active imitation learning methods solicit expert interventions to address the limitations. However, recent active imitation learning methods are designed based on human intuitions or empirical experience without theoretical guarantee. In this paper, we propose a novel active imitation learning framework based on a teacher-student interaction model, in which the teacher's goal is to identify the best teaching behavior and actively affect the student's learning process. By solving the optimization objective of this framework, we propose a practical implementation, naming it AdapMen. Theoretical analysis shows that AdapMen can improve the error bound and avoid compounding error under mild conditions. Experiments on the MetaDrive benchmark and Atari 2600 games validate our theoretical analysis and show that our method achieves near-expert performance with much less expert involvement and total sampling steps than previous methods. The code is available at https://github.com/liuxhym/AdapMen.

Yi-Chen Li, Fuxiang Zhang, Wenjie Qiu et al.

Large Language Models (LLMs), trained on a large amount of corpus, have demonstrated remarkable abilities. However, it may not be sufficient to directly apply open-source LLMs like Llama to certain real-world scenarios, since most of them are trained for \emph{general} purposes. Thus, the demands for customizing publicly available LLMs emerge, but are currently under-studied. In this work, we consider customizing pre-trained LLMs with new human preferences. Specifically, the LLM should not only meet the new preference but also preserve its original capabilities after customization. Drawing inspiration from the observation that human preference can be expressed as a reward model, we propose to cast LLM customization as optimizing the sum of two reward functions, one of which (denoted as $r_1$) was used to pre-train the LLM while the other (denoted as $r_2$) characterizes the new human preference. The obstacle here is that both reward functions are unknown, making the application of modern reinforcement learning methods infeasible. Thanks to the residual Q-learning framework, we can restore the customized LLM with the pre-trained LLM and the \emph{residual Q-function} without the reward function $r_1$. Moreover, we find that for a fixed pre-trained LLM, the reward function $r_2$ can be derived from the residual Q-function, enabling us to directly learn the residual Q-function from the new human preference data upon the Bradley-Terry model. We name our method Q-Adapter as it introduces an adapter module to approximate the residual Q-function for customizing the pre-trained LLM towards the new preference. Experiments based on the Llama-3.1 model on the DSP dataset and HH-RLHF dataset illustrate the superior effectiveness of Q-Adapter on both retaining existing knowledge and learning new preferences. Code is available at https://github.com/mansicer/Q-Adapter.

Cong Guan, Feng Chen, Lei Yuan et al.

Utilizing messages from teammates can improve coordination in cooperative Multi-agent Reinforcement Learning (MARL). Previous works typically combine raw messages of teammates with local information as inputs for policy. However, neglecting message aggregation poses significant inefficiency for policy learning. Motivated by recent advances in representation learning, we argue that efficient message aggregation is essential for good coordination in cooperative MARL. In this paper, we propose Multi-Agent communication via Self-supervised Information Aggregation (MASIA), where agents can aggregate the received messages into compact representations with high relevance to augment the local policy. Specifically, we design a permutation invariant message encoder to generate common information-aggregated representation from messages and optimize it via reconstructing and shooting future information in a self-supervised manner. Hence, each agent would utilize the most relevant parts of the aggregated representation for decision-making by a novel message extraction mechanism. Furthermore, considering the potential of offline learning for real-world applications, we build offline benchmarks for multi-agent communication, which is the first as we know. Empirical results demonstrate the superiority of our method in both online and offline settings. We also release the built offline benchmarks in this paper as a testbed for communication ability validation to facilitate further future research.

Ke Xue, Jiacheng Xu, Lei Yuan et al.

Automated algorithm configuration relieves users from tedious, trial-and-error tuning tasks. A popular algorithm configuration tuning paradigm is dynamic algorithm configuration (DAC), in which an agent learns dynamic configuration policies across instances by reinforcement learning (RL). However, in many complex algorithms, there may exist different types of configuration hyperparameters, and such heterogeneity may bring difficulties for classic DAC which uses a single-agent RL policy. In this paper, we aim to address this issue and propose multi-agent DAC (MA-DAC), with one agent working for one type of configuration hyperparameter. MA-DAC formulates the dynamic configuration of a complex algorithm with multiple types of hyperparameters as a contextual multi-agent Markov decision process and solves it by a cooperative multi-agent RL (MARL) algorithm. To instantiate, we apply MA-DAC to a well-known optimization algorithm for multi-objective optimization problems. Experimental results show the effectiveness of MA-DAC in not only achieving superior performance compared with other configuration tuning approaches based on heuristic rules, multi-armed bandits, and single-agent RL, but also being capable of generalizing to different problem classes. Furthermore, we release the environments in this paper as a benchmark for testing MARL algorithms, with the hope of facilitating the application of MARL.

Tao Jiang, Zichuan Lin, Lihe Li et al.

Large transformer models, trained on diverse datasets, have demonstrated impressive few-shot performance on previously unseen tasks without requiring parameter updates. This capability has also been explored in Reinforcement Learning (RL), where agents interact with the environment to retrieve context and maximize cumulative rewards, showcasing strong adaptability in complex settings. However, in cooperative Multi-Agent Reinforcement Learning (MARL), where agents must coordinate toward a shared goal, decentralized policy deployment can lead to mismatches in task alignment and reward assignment, limiting the efficiency of policy adaptation. To address this challenge, we introduce Multi-agent In-context Coordination via Decentralized Memory Retrieval (MAICC), a novel approach designed to enhance coordination by fast adaptation. Our method involves training a centralized embedding model to capture fine-grained trajectory representations, followed by decentralized models that approximate the centralized one to obtain team-level task information. Based on the learned embeddings, relevant trajectories are retrieved as context, which, combined with the agents' current sub-trajectories, inform decision-making. During decentralized execution, we introduce a novel memory mechanism that effectively balances test-time online data with offline memory. Based on the constructed memory, we propose a hybrid utility score that incorporates both individual- and team-level returns, ensuring credit assignment across agents. Extensive experiments on cooperative MARL benchmarks, including Level-Based Foraging (LBF) and SMAC (v1/v2), show that MAICC enables faster adaptation to unseen tasks compared to existing methods. Code is available at https://github.com/LAMDA-RL/MAICC.

Chen-Xiao Gao, Shengjun Fang, Chenjun Xiao et al.

Offline preference-based reinforcement learning (RL), which focuses on optimizing policies using human preferences between pairs of trajectory segments selected from an offline dataset, has emerged as a practical avenue for RL applications. Existing works rely on extracting step-wise reward signals from trajectory-wise preference annotations, assuming that preferences correlate with the cumulative Markovian rewards. However, such methods fail to capture the holistic perspective of data annotation: Humans often assess the desirability of a sequence of actions by considering the overall outcome rather than the immediate rewards. To address this challenge, we propose to model human preferences using rewards conditioned on future outcomes of the trajectory segments, i.e. the hindsight information. For downstream RL optimization, the reward of each step is calculated by marginalizing over possible future outcomes, the distribution of which is approximated by a variational auto-encoder trained using the offline dataset. Our proposed method, Hindsight Preference Learning (HPL), can facilitate credit assignment by taking full advantage of vast trajectory data available in massive unlabeled datasets. Comprehensive empirical studies demonstrate the benefits of HPL in delivering robust and advantageous rewards across various domains. Our code is publicly released at https://github.com/typoverflow/WiseRL.

Fuxiang Zhang, Junyou Li, Yi-Chen Li et al.

Low sample efficiency is an enduring challenge of reinforcement learning (RL). With the advent of versatile large language models (LLMs), recent works impart common-sense knowledge to accelerate policy learning for RL processes. However, we note that such guidance is often tailored for one specific task but loses generalizability. In this paper, we introduce a framework that harnesses LLMs to extract background knowledge of an environment, which contains general understandings of the entire environment, making various downstream RL tasks benefit from one-time knowledge representation. We ground LLMs by feeding a few pre-collected experiences and requesting them to delineate background knowledge of the environment. Afterward, we represent the output knowledge as potential functions for potential-based reward shaping, which has a good property for maintaining policy optimality from task rewards. We instantiate three variants to prompt LLMs for background knowledge, including writing code, annotating preferences, and assigning goals. Our experiments show that these methods achieve significant sample efficiency improvements in a spectrum of downstream tasks from Minigrid and Crafter domains.

Xiong-Hui Chen, Junyin Ye, Hang Zhao et al.

Imitation learning (IL) enables agents to mimic expert behaviors. Most previous IL techniques focus on precisely imitating one policy through mass demonstrations. However, in many applications, what humans require is the ability to perform various tasks directly through a few demonstrations of corresponding tasks, where the agent would meet many unexpected changes when deployed. In this scenario, the agent is expected to not only imitate the demonstration but also adapt to unforeseen environmental changes. This motivates us to propose a new topic called imitator learning (ItorL), which aims to derive an imitator module that can on-the-fly reconstruct the imitation policies based on very limited expert demonstrations for different unseen tasks, without any extra adjustment. In this work, we focus on imitator learning based on only one expert demonstration. To solve ItorL, we propose Demo-Attention Actor-Critic (DAAC), which integrates IL into a reinforcement-learning paradigm that can regularize policies' behaviors in unexpected situations. Besides, for autonomous imitation policy building, we design a demonstration-based attention architecture for imitator policy that can effectively output imitated actions by adaptively tracing the suitable states in demonstrations. We develop a new navigation benchmark and a robot environment for \topic~and show that DAAC~outperforms previous imitation methods \textit{with large margins} both on seen and unseen tasks.

Chao Chen, Dawei Wang, Feng Mao et al.

Semi-supervised Anomaly Detection (AD) is a kind of data mining task which aims at learning features from partially-labeled datasets to help detect outliers. In this paper, we classify existing semi-supervised AD methods into two categories: unsupervised-based and supervised-based, and point out that most of them suffer from insufficient exploitation of labeled data and under-exploration of unlabeled data. To tackle these problems, we propose Deep Anomaly Detection and Search (DADS), which applies Reinforcement Learning (RL) to balance exploitation and exploration. During the training process, the agent searches for possible anomalies with hierarchically-structured datasets and uses the searched anomalies to enhance performance, which in essence draws lessons from the idea of ensemble learning. Experimentally, we compare DADS with several state-of-the-art methods in the settings of leveraging labeled known anomalies to detect both other known anomalies and unknown anomalies. Results show that DADS can efficiently and precisely search anomalies from unlabeled data and learn from them, thus achieving good performance.

Chengxing Jia, Hao Yin, Chenxiao Gao et al.

Model-based offline optimization with dynamics-aware policy provides a new perspective for policy learning and out-of-distribution generalization, where the learned policy could adapt to different dynamics enumerated at the training stage. But due to the limitation under the offline setting, the learned model could not mimic real dynamics well enough to support reliable out-of-distribution exploration, which still hinders policy to generalize well. To narrow the gap, previous works roughly ensemble randomly initialized models to better approximate the real dynamics. However, such practice is costly and inefficient, and provides no guarantee on how well the real dynamics could be approximated by the learned models, which we name coverability in this paper. We actively address this issue by generating models with provable ability to cover real dynamics in an efficient and controllable way. To that end, we design a distance metric for dynamic models based on the occupancy of policies under the dynamics, and propose an algorithm to generate models optimizing their coverage for the real dynamics. We give a theoretical analysis on the model generation process and proves that our algorithm could provide enhanced coverability. As a downstream task, we train a dynamics-aware policy with minor or no conservative penalty, and experiments demonstrate that our algorithm outperforms prior offline methods on existing offline RL benchmarks. We also discover that policies learned by our method have better zero-shot transfer performance, implying their better generalization.

Jiafei Lyu, Kang Xu, Jiacheng Xu et al.

We consider off-dynamics reinforcement learning (RL) where one needs to transfer policies across different domains with dynamics mismatch. Despite the focus on developing dynamics-aware algorithms, this field is hindered due to the lack of a standard benchmark. To bridge this gap, we introduce ODRL, the first benchmark tailored for evaluating off-dynamics RL methods. ODRL contains four experimental settings where the source and target domains can be either online or offline, and provides diverse tasks and a broad spectrum of dynamics shifts, making it a reliable platform to comprehensively evaluate the agent's adaptation ability to the target domain. Furthermore, ODRL includes recent off-dynamics RL algorithms in a unified framework and introduces some extra baselines for different settings, all implemented in a single-file manner. To unpack the true adaptation capability of existing methods, we conduct extensive benchmarking experiments, which show that no method has universal advantages across varied dynamics shifts. We hope this benchmark can serve as a cornerstone for future research endeavors. Our code is publicly available at https://github.com/OffDynamicsRL/off-dynamics-rl.

Xuqin Zhang, Quan He, Zhenrui Zheng et al.

Reinforcement learning (RL) has emerged as a dominant paradigm for eliciting long-horizon reasoning in Large Language Models (LLMs). However, scaling Tool-Integrated Reasoning (TIR) via RL remains challenging due to interaction collapse: a pathological state where models fail to sustain multi-turn tool usage, instead degenerating into heavy internal reasoning with only trivial, post-hoc code verification. We systematically study three questions: (i) how cold-start SFT induces an agentic, tool-using behavioral prior, (ii) how the interaction density of cold-start trajectories shapes exploration and downstream RL outcomes, and (iii) how the RL interaction budget affects learning dynamics and generalization under varying inference-time budgets. We then introduce ASTER (Agentic Scaling with Tool-integrated Extended Reasoning), a framework that circumvents this collapse through a targeted cold-start strategy prioritizing interaction-dense trajectories. We find that a small expert cold-start set of just 4K interaction-dense trajectories yields the strongest downstream performance, establishing a robust prior that enables superior exploration during extended RL training. Extensive evaluations demonstrate that ASTER-4B achieves state-of-the-art results on competitive mathematical benchmarks, reaching 90.0% on AIME 2025, surpassing leading frontier open-source models, including DeepSeek-V3.2-Exp.

Feng Xu, Yan Yin, Xinyu Zhang et al.

Alphas are pivotal in providing signals for quantitative trading. The industry highly values the discovery of formulaic alphas for their interpretability and ease of analysis, compared with the expressive yet overfitting-prone black-box alphas. In this work, we focus on discovering formulaic alphas. Prior studies on automatically generating a collection of formulaic alphas were mostly based on genetic programming (GP), which is known to suffer from the problems of being sensitive to the initial population, converting to local optima, and slow computation speed. Recent efforts employing deep reinforcement learning (DRL) for alpha discovery have not fully addressed key practical considerations such as alpha correlations and validity, which are crucial for their effectiveness. In this work, we propose a novel framework for alpha discovery using DRL by formulating the alpha discovery process as program construction. Our agent, $\text{Alpha}^2$, assembles an alpha program optimized for an evaluation metric. A search algorithm guided by DRL navigates through the search space based on value estimates for potential alpha outcomes. The evaluation metric encourages both the performance and the diversity of alphas for a better final trading strategy. Our formulation of searching alphas also brings the advantage of pre-calculation dimensional analysis, ensuring the logical soundness of alphas, and pruning the vast search space to a large extent. Empirical experiments on real-world stock markets demonstrates $\text{Alpha}^2$'s capability to identify a diverse set of logical and effective alphas, which significantly improves the performance of the final trading strategy. The code of our method is available at https://github.com/x35f/alpha2.

Renzhe Zhou, Chen-Xiao Gao, Zongzhang Zhang et al.

Generalization and sample efficiency have been long-standing issues concerning reinforcement learning, and thus the field of Offline Meta-Reinforcement Learning~(OMRL) has gained increasing attention due to its potential of solving a wide range of problems with static and limited offline data. Existing OMRL methods often assume sufficient training tasks and data coverage to apply contrastive learning to extract task representations. However, such assumptions are not applicable in several real-world applications and thus undermine the generalization ability of the representations. In this paper, we consider OMRL with two types of data limitations: limited training tasks and limited behavior diversity and propose a novel algorithm called GENTLE for learning generalizable task representations in the face of data limitations. GENTLE employs Task Auto-Encoder~(TAE), which is an encoder-decoder architecture to extract the characteristics of the tasks. Unlike existing methods, TAE is optimized solely by reconstruction of the state transition and reward, which captures the generative structure of the task models and produces generalizable representations when training tasks are limited. To alleviate the effect of limited behavior diversity, we consistently construct pseudo-transitions to align the data distribution used to train TAE with the data distribution encountered during testing. Empirically, GENTLE significantly outperforms existing OMRL methods on both in-distribution tasks and out-of-distribution tasks across both the given-context protocol and the one-shot protocol.

Lei Song, Chenxiao Gao, Ke Xue et al.

Black-Box Optimization (BBO) has found successful applications in many fields of science and engineering. Recently, there has been a growing interest in meta-learning particular components of BBO algorithms to speed up optimization and get rid of tedious hand-crafted heuristics. As an extension, learning the entire algorithm from data requires the least labor from experts and can provide the most flexibility. In this paper, we propose RIBBO, a method to reinforce-learn a BBO algorithm from offline data in an end-to-end fashion. RIBBO employs expressive sequence models to learn the optimization histories produced by multiple behavior algorithms and tasks, leveraging the in-context learning ability of large models to extract task information and make decisions accordingly. Central to our method is to augment the optimization histories with \textit{regret-to-go} tokens, which are designed to represent the performance of an algorithm based on cumulative regret over the future part of the histories. The integration of regret-to-go tokens enables RIBBO to automatically generate sequences of query points that satisfy the user-desired regret, which is verified by its universally good empirical performance on diverse problems, including BBO benchmark functions, hyper-parameter optimization and robot control problems.

Rui Yu, Shenghua Wan, Yucen Wang et al.

In reinforcement learning (RL), agents continually interact with the environment and use the feedback to refine their behavior. To guide policy optimization, reward models are introduced as proxies of the desired objectives, such that when the agent maximizes the accumulated reward, it also fulfills the task designer's intentions. Recently, significant attention from both academic and industrial researchers has focused on developing reward models that not only align closely with the true objectives but also facilitate policy optimization. In this survey, we provide a comprehensive review of reward modeling techniques within the deep RL literature. We begin by outlining the background and preliminaries in reward modeling. Next, we present an overview of recent reward modeling approaches, categorizing them based on the source, the mechanism, and the learning paradigm. Building on this understanding, we discuss various applications of these reward modeling techniques and review methods for evaluating reward models. Finally, we conclude by highlighting promising research directions in reward modeling. Altogether, this survey includes both established and emerging methods, filling the vacancy of a systematic review of reward models in current literature.

Chen-Xiao Gao, Chenyang Wu, Mingjun Cao et al.

Behavior regularization, which constrains the policy to stay close to some behavior policy, is widely used in offline reinforcement learning (RL) to manage the risk of hazardous exploitation of unseen actions. Nevertheless, existing literature on behavior-regularized RL primarily focuses on explicit policy parameterizations, such as Gaussian policies. Consequently, it remains unclear how to extend this framework to more advanced policy parameterizations, such as diffusion models. In this paper, we introduce BDPO, a principled behavior-regularized RL framework tailored for diffusion-based policies, thereby combining the expressive power of diffusion policies and the robustness provided by regularization. The key ingredient of our method is to calculate the Kullback-Leibler (KL) regularization analytically as the accumulated discrepancies in reverse-time transition kernels along the diffusion trajectory. By integrating the regularization, we develop an efficient two-time-scale actor-critic RL algorithm that produces the optimal policy while respecting the behavior constraint. Comprehensive evaluations conducted on synthetic 2D tasks and continuous control tasks from the D4RL benchmark validate its effectiveness and superior performance.

Chengxing Jia, Fuxiang Zhang, Yi-Chen Li et al.

Offline meta-reinforcement learning (OMRL) proficiently allows an agent to tackle novel tasks while solely relying on a static dataset. For precise and efficient task identification, existing OMRL research suggests learning separate task representations that be incorporated with policy input, thus forming a context-based meta-policy. A major approach to train task representations is to adopt contrastive learning using multi-task offline data. The dataset typically encompasses interactions from various policies (i.e., the behavior policies), thus providing a plethora of contextual information regarding different tasks. Nonetheless, amassing data from a substantial number of policies is not only impractical but also often unattainable in realistic settings. Instead, we resort to a more constrained yet practical scenario, where multi-task data collection occurs with a limited number of policies. We observed that learned task representations from previous OMRL methods tend to correlate spuriously with the behavior policy instead of reflecting the essential characteristics of the task, resulting in unfavorable out-of-distribution generalization. To alleviate this issue, we introduce a novel algorithm to disentangle the impact of behavior policy from task representation learning through a process called adversarial data augmentation. Specifically, the objective of adversarial data augmentation is not merely to generate data analogous to offline data distribution; instead, it aims to create adversarial examples designed to confound learned task representations and lead to incorrect task identification. Our experiments show that learning from such adversarial samples significantly enhances the robustness and effectiveness of the task identification process and realizes satisfactory out-of-distribution generalization.

Feng Chen, Fuguang Han, Cong Guan et al.

Given the inherent non-stationarity prevalent in real-world applications, continual Reinforcement Learning (RL) aims to equip the agent with the capability to address a series of sequentially presented decision-making tasks. Within this problem setting, a pivotal challenge revolves around \textit{catastrophic forgetting} issue, wherein the agent is prone to effortlessly erode the decisional knowledge associated with past encountered tasks when learning the new one. In recent progresses, the \textit{generative replay} methods have showcased substantial potential by employing generative models to replay data distribution of past tasks. Compared to storing the data from past tasks directly, this category of methods circumvents the growing storage overhead and possible data privacy concerns. However, constrained by the expressive capacity of generative models, existing \textit{generative replay} methods face challenges in faithfully reconstructing the data distribution of past tasks, particularly in scenarios with a myriad of tasks or high-dimensional data. Inspired by the success of diffusion models in various generative tasks, this paper introduces a novel continual RL algorithm DISTR (Diffusion-based Trajectory Replay) that employs a diffusion model to memorize the high-return trajectory distribution of each encountered task and wakeups these distributions during the policy learning on new tasks. Besides, considering the impracticality of replaying all past data each time, a prioritization mechanism is proposed to prioritize the trajectory replay of pivotal tasks in our method. Empirical experiments on the popular continual RL benchmark \texttt{Continual World} demonstrate that our proposed method obtains a favorable balance between \textit{stability} and \textit{plasticity}, surpassing various existing continual RL baselines in average success rate.

Wenjie Qiu, Yi-Chen Li, Xuqin Zhang et al.

Learning reward models from human preference datasets and subsequently optimizing language models via reinforcement learning has emerged as a fundamental paradigm for aligning LLMs with human preferences. The performance of the reward model plays a crucial role in the effectiveness of alignment. Previous reward models operate at a coarse-grained level, requiring the generation of a complete response to obtain a reward value. The sparse reward may present challenges for downstream reinforcement learning. While recent efforts have attempted to learn token-level reward models, the lack of explicit semantic information makes it difficult to model the credit of every individual token. In this paper, we propose assigning scores to every sentence, introducing an intermediate-grained reward model. By segmenting the complete response into sentences and applying differential operations to reward output at the start and end positions of each sentence, we can effectively model the rewards of sentences. Moreover, a novel attention mechanism is introduced to aggregate the scores of all sentences into a response-level score, which allows it to be trained using the Bradley-Terry model. On common benchmarks, our method outperforms the response-level reward model by 2.7% on RewardBench (for reward modeling evaluation) and surpasses all baselines on AlpacaEval (for alignment evaluation).

Shihan Dou, Ming Zhang, Chenhao Huang et al.

We introduce EvaLearn, a pioneering benchmark designed to evaluate large language models (LLMs) on their learning capability and efficiency in challenging tasks, a critical, yet underexplored aspect of model potential. EvaLearn contains 648 challenging problems across six task types, grouped into 182 sequences, each sequence dedicated to one task type. Diverging from most existing benchmarks that evaluate models in parallel, EvaLearn requires models to solve problems sequentially, allowing them to leverage the experience gained from previous solutions. EvaLearn provides five comprehensive automated metrics to evaluate models and quantify their learning capability and efficiency. We extensively benchmark nine frontier models and observe varied performance profiles: some models, such as Claude-3.7-sonnet, start with moderate initial performance but exhibit strong learning ability, while some models struggle to benefit from experience and may even show negative transfer. Moreover, we investigate model performance under two learning settings and find that instance-level rubrics and teacher-model feedback further facilitate model learning. Importantly, we observe that current LLMs with stronger static abilities do not show a clear advantage in learning capability across all tasks, highlighting that EvaLearn evaluates a new dimension of model performance. We hope EvaLearn provides a novel evaluation perspective for assessing LLM potential and understanding the gap between models and human capabilities, promoting the development of deeper and more dynamic evaluation approaches. All datasets, the automatic evaluation framework, and the results studied in this paper are available at the GitHub repository.

Xinyu Zhang, Wenjie Qiu, Yi-Chen Li et al.

Developing policies that can adjust to non-stationary environments is essential for real-world reinforcement learning applications. However, learning such adaptable policies in offline settings, with only a limited set of pre-collected trajectories, presents significant challenges. A key difficulty arises because the limited offline data makes it hard for the context encoder to differentiate between changes in the environment dynamics and shifts in the behavior policy, often leading to context misassociations. To address this issue, we introduce a novel approach called Debiased Offline Representation for fast online Adaptation (DORA). DORA incorporates an information bottleneck principle that maximizes mutual information between the dynamics encoding and the environmental data, while minimizing mutual information between the dynamics encoding and the actions of the behavior policy. We present a practical implementation of DORA, leveraging tractable bounds of the information bottleneck principle. Our experimental evaluation across six benchmark MuJoCo tasks with variable parameters demonstrates that DORA not only achieves a more precise dynamics encoding but also significantly outperforms existing baselines in terms of performance.

Jing-Cheng Pang, Pengyuan Wang, Kaiyuan Li et al.

Large Language Models (LLMs) have exhibited remarkable performance across various natural language processing (NLP) tasks. However, fine-tuning these models often necessitates substantial supervision, which can be expensive and time-consuming to obtain. This paper introduces a novel unsupervised method called LanguageModel Self-Improvement by Reinforcement Learning Contemplation (SIRLC) that improves LLMs without reliance on external labels. Our approach is grounded in the observation that it is simpler for language models to assess text quality than to generate text. Building on this insight, SIRLC assigns LLMs dual roles as both student and teacher. As a student, the LLM generates answers to unlabeled questions, while as a teacher, it evaluates the generated text and assigns scores accordingly. The model parameters are updated using reinforcement learning to maximize the evaluation score. We demonstrate that SIRLC can be applied to various NLP tasks, such as reasoning problems, text generation, and machine translation. Our experiments show that SIRLC effectively improves LLM performance without external supervision, resulting in a 5.6% increase in answering accuracy for reasoning tasks and a rise in BERTScore from 0.82 to 0.86 for translation tasks. Furthermore, SIRLC can be applied to models of different sizes, showcasing its broad applicability.

Lei Yuan, Tao Jiang, Lihe Li et al.

Many multi-agent scenarios require message sharing among agents to promote coordination, hastening the robustness of multi-agent communication when policies are deployed in a message perturbation environment. Major relevant works tackle this issue under specific assumptions, like a limited number of message channels would sustain perturbations, limiting the efficiency in complex scenarios. In this paper, we take a further step addressing this issue by learning a robust multi-agent communication policy via multi-view message certification, dubbed CroMAC. Agents trained under CroMAC can obtain guaranteed lower bounds on state-action values to identify and choose the optimal action under a worst-case deviation when the received messages are perturbed. Concretely, we first model multi-agent communication as a multi-view problem, where every message stands for a view of the state. Then we extract a certificated joint message representation by a multi-view variational autoencoder (MVAE) that uses a product-of-experts inference network. For the optimization phase, we do perturbations in the latent space of the state for a certificate guarantee. Then the learned joint message representation is used to approximate the certificated state representation during training. Extensive experiments in several cooperative multi-agent benchmarks validate the effectiveness of the proposed CroMAC.

Cong Fei, Bin Wang, Yuzheng Zhuang et al.

Generative adversarial imitation learning (GAIL) has shown promising results by taking advantage of generative adversarial nets, especially in the field of robot learning. However, the requirement of isolated single modal demonstrations limits the scalability of the approach to real world scenarios such as autonomous vehicles' demand for a proper understanding of human drivers' behavior. In this paper, we propose a novel multi-modal GAIL framework, named Triple-GAIL, that is able to learn skill selection and imitation jointly from both expert demonstrations and continuously generated experiences with data augmentation purpose by introducing an auxiliary skill selector. We provide theoretical guarantees on the convergence to optima for both of the generator and the selector respectively. Experiments on real driver trajectories and real-time strategy game datasets demonstrate that Triple-GAIL can better fit multi-modal behaviors close to the demonstrators and outperforms state-of-the-art methods.

Tianpei Yang, Jianye Hao, Zhaopeng Meng et al.

Transfer Learning (TL) has shown great potential to accelerate Reinforcement Learning (RL) by leveraging prior knowledge from past learned policies of relevant tasks. Existing transfer approaches either explicitly computes the similarity between tasks or select appropriate source policies to provide guided explorations for the target task. However, how to directly optimize the target policy by alternatively utilizing knowledge from appropriate source policies without explicitly measuring the similarity is currently missing. In this paper, we propose a novel Policy Transfer Framework (PTF) to accelerate RL by taking advantage of this idea. Our framework learns when and which source policy is the best to reuse for the target policy and when to terminate it by modeling multi-policy transfer as the option learning problem. PTF can be easily combined with existing deep RL approaches. Experimental results show it significantly accelerates the learning process and surpasses state-of-the-art policy transfer methods in terms of learning efficiency and final performance in both discrete and continuous action spaces.

Xiaobai Ma, Katherine Driggs-Campbell, Zongzhang Zhang et al.

Gradient-based methods are often used for policy optimization in deep reinforcement learning, despite being vulnerable to local optima and saddle points. Although gradient-free methods (e.g., genetic algorithms or evolution strategies) help mitigate these issues, poor initialization and local optima are still concerns in highly nonconvex spaces. This paper presents a method for policy optimization based on Monte-Carlo tree search and gradient-free optimization. Our method, called Monte-Carlo tree search for policy optimization (MCTSPO), provides a better exploration-exploitation trade-off through the use of the upper confidence bound heuristic. We demonstrate improved performance on reinforcement learning tasks with deceptive or sparse reward functions compared to popular gradient-based and deep genetic algorithm baselines.

Hongyao Tang, Jianye Hao, Tangjie Lv et al.

Multiagent reinforcement learning (MARL) is commonly considered to suffer from non-stationary environments and exponentially increasing policy space. It would be even more challenging when rewards are sparse and delayed over long trajectories. In this paper, we study hierarchical deep MARL in cooperative multiagent problems with sparse and delayed reward. With temporal abstraction, we decompose the problem into a hierarchy of different time scales and investigate how agents can learn high-level coordination based on the independent skills learned at the low level. Three hierarchical deep MARL architectures are proposed to learn hierarchical policies under different MARL paradigms. Besides, we propose a new experience replay mechanism to alleviate the issue of the sparse transitions at the high level of abstraction and the non-stationarity of multiagent learning. We empirically demonstrate the effectiveness of our approaches in two domains with extremely sparse feedback: (1) a variety of Multiagent Trash Collection tasks, and (2) a challenging online mobile game, i.e., Fever Basketball Defense.

Yan Zheng, Jianye Hao, Zongzhang Zhang

Recently, multiagent deep reinforcement learning (DRL) has received increasingly wide attention. Existing multiagent DRL algorithms are inefficient when facing with the non-stationarity due to agents update their policies simultaneously in stochastic cooperative environments. This paper extends the recently proposed weighted double estimator to the multiagent domain and propose a multiagent DRL framework, named weighted double deep Q-network (WDDQN). By utilizing the weighted double estimator and the deep neural network, WDDQN can not only reduce the bias effectively but also be extended to scenarios with raw visual inputs. To achieve efficient cooperation in the multiagent domain, we introduce the lenient reward network and the scheduled replay strategy. Experiments show that the WDDQN outperforms the existing DRL and multiaent DRL algorithms, i.e., double DQN and lenient Q-learning, in terms of the average reward and the convergence rate in stochastic cooperative environments.