Yi-Chen Li

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

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.

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.

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.

Yangyang Zhou, Yi-Chen Li

Reinforcement Learning from Human Feedback has become the standard paradigm for language model alignment, where reward models directly determine alignment effectiveness. In this work, we focus on how to evaluate the generalizability of reward models. By "generalizability", we mean the ability of RMs to correctly rank responses to align with diverse user preferences. However, existing reward model benchmarks are typically designed around a universal preference, failing to assess this generalization. To address this critical gap, we introduce RMGAP, a benchmark comprising 1,097 instances across Chat, Writing, Reasoning, and Safety domains. Since different users exhibit diverse preferences for the same task, we first generate four distinct responses with different linguistic profiles for each collected prompt. However, the original prompt set lacks the specificity to convey different preferences. We therefore construct tailored prompts by contrasting these candidates and designing scenarios in which one response becomes the uniquely appropriate choice. Moreover, we observe that users often express the same preference using different phrasings, and thus extend each prompt with two paraphrased variants. Our evaluation of 24 state-of-the-art RMs reveals their substantial limitations: even the best RM achieves only 49.27% Best-of-N accuracy, highlighting considerable room for improvement in reward model generalization. Related data and code are available at https://github.com/nanzhi84/RMGAP.

Tian Xu, Chenyang Wang, Xiaochen Zhai et al.

Adversarial imitation learning (AIL) achieves high-quality imitation by mitigating compounding errors in behavioral cloning (BC), but often exhibits training instability due to adversarial optimization. To avoid this issue, a class of non-adversarial Q-based imitation learning (IL) methods, represented by IQ-Learn, has emerged and is widely believed to outperform BC by leveraging online environment interactions. However, this paper revisits IQ-Learn and demonstrates that it provably reduces to BC and suffers from an imitation gap lower bound with quadratic dependence on horizon, therefore still suffering from compounding errors. Theoretical analysis reveals that, despite using online interactions, IQ-Learn uniformly suppresses the Q-values for all actions on states uncovered by demonstrations, thereby failing to generalize. To address this limitation, we introduce a primal-dual framework for distribution matching, yielding a new Q-based IL method, Dual Q-DM. The key mechanism in Dual Q-DM is incorporating Bellman constraints to propagate high Q-values from visited states to unvisited ones, thereby achieving generalization beyond demonstrations. We prove that Dual Q-DM is equivalent to AIL and can recover expert actions beyond demonstrations, thereby mitigating compounding errors. To the best of our knowledge, Dual Q-DM is the first non-adversarial IL method that is theoretically guaranteed to eliminate compounding errors. Experimental results further corroborate our theoretical results.

Peng-Yuan Wang, Ziniu Li, Tian Xu et al.

Improving data utilization efficiency is critical for scaling reinforcement learning (RL) for long-horizon tasks where generating trajectories is expensive. However, the dominant RL methods for LLMs are largely on-policy: they update each batch of data only once, discard it, and then collect fresh samples, resulting in poor sample efficiency. In this work, we explore an alternative value-based RL framework for LLMs that naturally enables off-policy learning. We propose ReVal, a Bellman-update-based method that combines stepwise signals capturing internal consistency with trajectory-level signals derived from outcome verification. ReVal naturally supports replay-buffer-based training, allowing efficient reuse of past trajectories. Experiments on standard mathematical reasoning benchmarks show that ReVal not only converges faster but also outperforms GRPO in final performance. On DeepSeek-R1-Distill-1.5B, ReVal improves training efficiency and achieves improvement of 2.7% in AIME24 and 4.5% in out-of-domain benchmark GPQA over GRPO. These results suggest that value-based RL is a practical alternative to policy-based methods for LLM training.

Yi-Chen Li, Tian Xu, Yang Yu et al.

The alignment of Large Language Models (LLMs) is critically dependent on reward models trained on costly human preference data. While recent work explores bypassing this cost with AI feedback, these methods often lack a rigorous theoretical foundation. In this paper, we discover that a powerful generalist reward model is already latently present within any LLM trained via standard next-token prediction. We prove that this endogenous reward is not a heuristic, but is theoretically equivalent to a reward function learned through offline inverse reinforcement learning. This connection allows us to directly elicit a high-quality reward signal from a base (pre-trained or supervised fine-tuned) model without any further training. Critically, we also prove that subsequent reinforcement learning using this endogenous reward leads to a policy with a provably superior error bound compared to the base model. To our best knowledge, this is the first theoretical proof of the effectiveness of reinforcement learning for LLMs. Our experiments validate this theory, demonstrating that our method not only outperforms existing LLM-as-a-judge approaches but can also surpass explicitly trained reward models. These findings suggest that the reward modeling stage can be replaced by a principled method of eliciting the knowledge already captured during pre-training, heralding a more efficient, powerful, and scalable paradigm for LLMs alignment as well as multi-modal models.

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.

Chengxing Jia, Ziniu Li, Pengyuan Wang et al.

Adapting Large Language Models (LLMs) to downstream tasks using Reinforcement Learning (RL) has proven to be an effective approach. However, LLMs do not inherently define the structure of an agent for RL training, particularly in terms of defining the action space. This paper studies learning a compact latent action space to enhance the controllability and exploration of RL for LLMs. We propose Controlling Large Language Models with Latent Actions (CoLA), a framework that integrates a latent action space into pre-trained LLMs. We apply CoLA to the Llama-3.1-8B model. Our experiments demonstrate that, compared to RL with token-level actions, CoLA's latent action enables greater semantic diversity in text generation. For enhancing downstream tasks, we show that CoLA with RL achieves a score of 42.4 on the math500 benchmark, surpassing the baseline score of 38.2, and reaches 68.2 when augmented with a Monte Carlo Tree Search variant. Furthermore, CoLA with RL consistently improves performance on agent-based tasks without degrading the pre-trained LLM's capabilities, unlike the baseline. Finally, CoLA reduces computation time by half in tasks involving enhanced thinking prompts for LLMs by RL. These results highlight CoLA's potential to advance RL-based adaptation of LLMs for downstream applications.

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).

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.