Qianchuan Zhao

Xiaoteng Ma, Zhipeng Liang, Jose Blanchet et al. · tsinghua

Among the reasons hindering reinforcement learning (RL) applications to real-world problems, two factors are critical: limited data and the mismatch between the testing environment (real environment in which the policy is deployed) and the training environment (e.g., a simulator). This paper attempts to address these issues simultaneously with distributionally robust offline RL, where we learn a distributionally robust policy using historical data obtained from the source environment by optimizing against a worst-case perturbation thereof. In particular, we move beyond tabular settings and consider linear function approximation. More specifically, we consider two settings, one where the dataset is well-explored and the other where the dataset has sufficient coverage of the optimal policy. We propose two algorithms~-- one for each of the two settings~-- that achieve error bounds $\tilde{O}(d^{1/2}/N^{1/2})$ and $\tilde{O}(d^{3/2}/N^{1/2})$ respectively, where $d$ is the dimension in the linear function approximation and $N$ is the number of trajectories in the dataset. To the best of our knowledge, they provide the first non-asymptotic results of the sample complexity in this setting. Diverse experiments are conducted to demonstrate our theoretical findings, showing the superiority of our algorithm against the non-robust one.

Chenghao Li, Tonghan Wang, Chongjie Zhang et al. · harvard, tsinghua

In the realm of multi-agent reinforcement learning, intrinsic motivations have emerged as a pivotal tool for exploration. While the computation of many intrinsic rewards relies on estimating variational posteriors using neural network approximators, a notable challenge has surfaced due to the limited expressive capability of these neural statistics approximators. We pinpoint this challenge as the "revisitation" issue, where agents recurrently explore confined areas of the task space. To combat this, we propose a dynamic reward scaling approach. This method is crafted to stabilize the significant fluctuations in intrinsic rewards in previously explored areas and promote broader exploration, effectively curbing the revisitation phenomenon. Our experimental findings underscore the efficacy of our approach, showcasing enhanced performance in demanding environments like Google Research Football and StarCraft II micromanagement tasks, especially in sparse reward settings.

Xiaoteng Ma, Shuai Ma, Li Xia et al. · tsinghua

Keeping risk under control is often more crucial than maximizing expected rewards in real-world decision-making situations, such as finance, robotics, autonomous driving, etc. The most natural choice of risk measures is variance, which penalizes the upside volatility as much as the downside part. Instead, the (downside) semivariance, which captures the negative deviation of a random variable under its mean, is more suitable for risk-averse proposes. This paper aims at optimizing the mean-semivariance (MSV) criterion in reinforcement learning w.r.t. steady reward distribution. Since semivariance is time-inconsistent and does not satisfy the standard Bellman equation, the traditional dynamic programming methods are inapplicable to MSV problems directly. To tackle this challenge, we resort to Perturbation Analysis (PA) theory and establish the performance difference formula for MSV. We reveal that the MSV problem can be solved by iteratively solving a sequence of RL problems with a policy-dependent reward function. Further, we propose two on-policy algorithms based on the policy gradient theory and the trust region method. Finally, we conduct diverse experiments from simple bandit problems to continuous control tasks in MuJoCo, which demonstrate the effectiveness of our proposed methods.

Hao Hu, Yiqin Yang, Qianchuan Zhao et al.

Offline reinforcement learning (RL) enables effective learning from previously collected data without exploration, which shows great promise in real-world applications when exploration is expensive or even infeasible. The discount factor, $γ$, plays a vital role in improving online RL sample efficiency and estimation accuracy, but the role of the discount factor in offline RL is not well explored. This paper examines two distinct effects of $γ$ in offline RL with theoretical analysis, namely the regularization effect and the pessimism effect. On the one hand, $γ$ is a regulator to trade-off optimality with sample efficiency upon existing offline techniques. On the other hand, lower guidance $γ$ can also be seen as a way of pessimism where we optimize the policy's performance in the worst possible models. We empirically verify the above theoretical observation with tabular MDPs and standard D4RL tasks. The results show that the discount factor plays an essential role in the performance of offline RL algorithms, both under small data regimes upon existing offline methods and in large data regimes without other conservative methods.

Yiqin Yang, Hao Hu, Wenzhe Li et al.

Offline reinforcement learning (RL) enables the agent to effectively learn from logged data, which significantly extends the applicability of RL algorithms in real-world scenarios where exploration can be expensive or unsafe. Previous works have shown that extracting primitive skills from the recurring and temporally extended structures in the logged data yields better learning. However, these methods suffer greatly when the primitives have limited representation ability to recover the original policy space, especially in offline settings. In this paper, we give a quantitative characterization of the performance of offline hierarchical learning and highlight the importance of learning lossless primitives. To this end, we propose to use a \emph{flow}-based structure as the representation for low-level policies. This allows us to represent the behaviors in the dataset faithfully while keeping the expression ability to recover the whole policy space. We show that such lossless primitives can drastically improve the performance of hierarchical policies. The experimental results and extensive ablation studies on the standard D4RL benchmark show that our method has a good representation ability for policies and achieves superior performance in most tasks.

Hao Hu, Yiqin Yang, Qianchuan Zhao et al.

Self-supervised methods have become crucial for advancing deep learning by leveraging data itself to reduce the need for expensive annotations. However, the question of how to conduct self-supervised offline reinforcement learning (RL) in a principled way remains unclear. In this paper, we address this issue by investigating the theoretical benefits of utilizing reward-free data in linear Markov Decision Processes (MDPs) within a semi-supervised setting. Further, we propose a novel, Provable Data Sharing algorithm (PDS) to utilize such reward-free data for offline RL. PDS uses additional penalties on the reward function learned from labeled data to prevent overestimation, ensuring a conservative algorithm. Our results on various offline RL tasks demonstrate that PDS significantly improves the performance of offline RL algorithms with reward-free data. Overall, our work provides a promising approach to leveraging the benefits of unlabeled data in offline RL while maintaining theoretical guarantees. We believe our findings will contribute to developing more robust self-supervised RL methods.

Chengqi Lyu, Songyang Gao, Yuzhe Gu et al.

Reasoning abilities, especially those for solving complex math problems, are crucial components of general intelligence. Recent advances by proprietary companies, such as o-series models of OpenAI, have made remarkable progress on reasoning tasks. However, the complete technical details remain unrevealed, and the techniques that are believed certainly to be adopted are only reinforcement learning (RL) and the long chain of thoughts. This paper proposes a new RL framework, termed OREAL, to pursue the performance limit that can be achieved through \textbf{O}utcome \textbf{RE}w\textbf{A}rd-based reinforcement \textbf{L}earning for mathematical reasoning tasks, where only binary outcome rewards are easily accessible. We theoretically prove that behavior cloning on positive trajectories from best-of-N (BoN) sampling is sufficient to learn the KL-regularized optimal policy in binary feedback environments. This formulation further implies that the rewards of negative samples should be reshaped to ensure the gradient consistency between positive and negative samples. To alleviate the long-existing difficulties brought by sparse rewards in RL, which are even exacerbated by the partial correctness of the long chain of thought for reasoning tasks, we further apply a token-level reward model to sample important tokens in reasoning trajectories for learning. With OREAL, for the first time, a 7B model can obtain 94.0 pass@1 accuracy on MATH-500 through RL, being on par with 32B models. OREAL-32B also surpasses previous 32B models trained by distillation with 95.0 pass@1 accuracy on MATH-500. Our investigation also indicates the importance of initial policy models and training queries for RL. Code, models, and data will be released to benefit future research\footnote{https://github.com/InternLM/OREAL}.

Hai-Jin Ding, Re-Bing Wu, Qian-Chuan Zhao

Cyber-security has become vital for modern networked control systems (NCS). In this paper, we propose that the emerging technology of quantum key distribution (QKD) can be applied to enhance the privacy and security of NCS up to an unbreakable level. QKD can continuously distribute random secret keys with much higher privacy between communication parties, and thus enable the one-time pad encryption that cannot be truly applied in classical networks. We show that the resulting overall security of NCS can be essentially improved, and present a composable definition of security based on the analysis of the key generation and management processes. Moreover, because the security is mainly determined by quantum keys rather than the complexity of encryption algorithms, the control performance can be improved as well by reducing the time delay using simpler algorithms. These advantages are demonstrated by the example of a remotely controlled servo system, showing that the introduction of QKD to NCS can simultaneously improve the security and performance by using the simplest encryption algorithm XOR. Furthermore, we propose a novel Kalman-filter embedded communication protocol that can more efficiently use the raw keys generated by QKD.

Dianyu Zhong, Yiqin Yang, Qianchuan Zhao

The large action space is one fundamental obstacle to deploying Reinforcement Learning methods in the real world. The numerous redundant actions will cause the agents to make repeated or invalid attempts, even leading to task failure. Although current algorithms conduct some initial explorations for this issue, they either suffer from rule-based systems or depend on expert demonstrations, which significantly limits their applicability in many real-world settings. In this work, we examine the theoretical analysis of what action can be eliminated in policy optimization and propose a novel redundant action filtering mechanism. Unlike other works, our method constructs the similarity factor by estimating the distance between the state distributions, which requires no prior knowledge. In addition, we combine the modified inverse model to avoid extensive computation in high-dimensional state space. We reveal the underlying structure of action spaces and propose a simple yet efficient redundant action filtering mechanism named No Prior Mask (NPM) based on the above techniques. We show the superior performance of our method by conducting extensive experiments on high-dimensional, pixel-input, and stochastic problems with various action redundancy. Our code is public online at https://github.com/zhongdy15/npm.

Runpeng Xie, Quanwei Wang, Hao Hu et al.

Comprehending natural language and following human instructions are critical capabilities for intelligent agents. However, the flexibility of linguistic instructions induces substantial ambiguity across language-conditioned tasks, severely degrading algorithmic performance. To address these limitations, we present a novel method named DAIL (Distributional Aligned Learning), featuring two key components: distributional policy and semantic alignment. Specifically, we provide theoretical results that the value distribution estimation mechanism enhances task differentiability. Meanwhile, the semantic alignment module captures the correspondence between trajectories and linguistic instructions. Extensive experimental results on both structured and visual observation benchmarks demonstrate that DAIL effectively resolves instruction ambiguities, achieving superior performance to baseline methods. Our implementation is available at https://github.com/RunpengXie/Distributional-Aligned-Learning.

Kailai Sun, Zhou Yang, Qianchuan Zhao

Depth images have a wide range of applications, such as 3D reconstruction, autonomous driving, augmented reality, robot navigation, and scene understanding. Commodity-grade depth cameras are hard to sense depth for bright, glossy, transparent, and distant surfaces. Although existing depth completion methods have achieved remarkable progress, their performance is limited when applied to complex indoor scenarios. To address these problems, we propose a two-step Transformer-based network for indoor depth completion. Unlike existing depth completion approaches, we adopt a self-supervision pre-training encoder based on the masked autoencoder to learn an effective latent representation for the missing depth value; then we propose a decoder based on a token fusion mechanism to complete (i.e., reconstruct) the full depth from the jointly RGB and incomplete depth image. Compared to the existing methods, our proposed network, achieves the state-of-the-art performance on the Matterport3D dataset. In addition, to validate the importance of the depth completion task, we apply our methods to indoor 3D reconstruction. The code, dataset, and demo are available at https://github.com/kailaisun/Indoor-Depth-Completion.

Yuhua Jiang, Qihan Liu, Xiaoteng Ma et al.

Among the great successes of Reinforcement Learning (RL), self-play algorithms play an essential role in solving competitive games. Current self-play algorithms optimize the agent to maximize expected win-rates against its current or historical copies, making it often stuck in the local optimum and its strategy style simple and homogeneous. A possible solution is to improve the diversity of policies, which helps the agent break the stalemate and enhances its robustness when facing different opponents. However, enhancing diversity in the self-play algorithms is not trivial. In this paper, we aim to introduce diversity from the perspective that agents could have diverse risk preferences in the face of uncertainty. Specifically, we design a novel reinforcement learning algorithm called Risk-sensitive Proximal Policy Optimization (RPPO), which smoothly interpolates between worst-case and best-case policy learning and allows for policy learning with desired risk preferences. Seamlessly integrating RPPO with population-based self-play, agents in the population optimize dynamic risk-sensitive objectives with experiences from playing against diverse opponents. Empirical results show that our method achieves comparable or superior performance in competitive games and that diverse modes of behaviors emerge. Our code is public online at \url{https://github.com/Jackory/RPBT}.

Kailai Sun, Xiaoteng Ma, Peng Liu et al.

Head detection in the indoor video is an essential component of building occupancy detection. While deep models have achieved remarkable progress in general object detection, they are not satisfying enough in complex indoor scenes. The indoor surveillance video often includes cluttered background objects, among which heads have small scales and diverse poses. In this paper, we propose Motion-aware Pseudo Siamese Network (MPSN), an end-to-end approach that leverages head motion information to guide the deep model to extract effective head features in indoor scenarios. By taking the pixel-wise difference of adjacent frames as the auxiliary input, MPSN effectively enhances human head motion information and removes the irrelevant objects in the background. Compared with prior methods, it achieves superior performance on the two indoor video datasets. Our experiments show that MPSN successfully suppresses static background objects and highlights the moving instances, especially human heads in indoor videos. We also compare different methods to capture head motion, which demonstrates the simplicity and flexibility of MPSN. To validate the robustness of MPSN, we conduct adversarial experiments with a mathematical solution of small perturbations for robust model selection. Finally, for confirming its potential in building control systems, we apply MPSN to occupancy counting. Code is available at https://github.com/pl-share/MPSN.

Yiqin Yang, Xiaoteng Ma, Chenghao Li et al.

Learning from datasets without interaction with environments (Offline Learning) is an essential step to apply Reinforcement Learning (RL) algorithms in real-world scenarios. However, compared with the single-agent counterpart, offline multi-agent RL introduces more agents with the larger state and action space, which is more challenging but attracts little attention. We demonstrate current offline RL algorithms are ineffective in multi-agent systems due to the accumulated extrapolation error. In this paper, we propose a novel offline RL algorithm, named Implicit Constraint Q-learning (ICQ), which effectively alleviates the extrapolation error by only trusting the state-action pairs given in the dataset for value estimation. Moreover, we extend ICQ to multi-agent tasks by decomposing the joint-policy under the implicit constraint. Experimental results demonstrate that the extrapolation error is successfully controlled within a reasonable range and insensitive to the number of agents. We further show that ICQ achieves the state-of-the-art performance in the challenging multi-agent offline tasks (StarCraft II). Our code is public online at https://github.com/YiqinYang/ICQ.

Yuhua Jiang, Yuwen Xiong, Yufeng Yuan et al.

Large Language Models (LLMs) have demonstrated impressive capabilities in complex reasoning tasks, yet they still struggle to reliably verify the correctness of their own outputs. Existing solutions to this verification challenge often depend on separate verifier models or require multi-stage self-correction training pipelines, which limit scalability. In this paper, we propose Policy as Generative Verifier (PAG), a simple and effective framework that empowers LLMs to self-correct by alternating between policy and verifier roles within a unified multi-turn reinforcement learning (RL) paradigm. Distinct from prior approaches that always generate a second attempt regardless of model confidence, PAG introduces a selective revision mechanism: the model revises its answer only when its own generative verification step detects an error. This verify-then-revise workflow not only alleviates model collapse but also jointly enhances both reasoning and verification abilities. Extensive experiments across diverse reasoning benchmarks highlight PAG's dual advancements: as a policy, it enhances direct generation and self-correction accuracy; as a verifier, its self-verification outperforms self-consistency.

Yuntao Shou, Peiqiang Yan, Xingjian Yuan et al.

In recent years, histopathological whole slide image (WSI)- based survival analysis has attracted much attention in medical image analysis. In practice, WSIs usually come from different hospitals or laboratories, which can be seen as different domains, and thus may have significant differences in imaging equipment, processing procedures, and sample sources. These differences generally result in large gaps in distribution between different WSI domains, and thus the survival analysis models trained on one domain may fail to transfer to another. To address this issue, we propose a Dual-branch Encoder and Two-level Alignment (DETA) framework to explore both feature and category-level alignment between different WSI domains. Specifically, we first formulate the concerned problem as graph domain adaptation (GDA) by virtue the graph representation of WSIs. Then we construct a dual-branch graph encoder, including the message passing branch and the shortest path branch, to explicitly and implicitly extract semantic information from the graph-represented WSIs. To realize GDA, we propose a two-level alignment approach: at the category level, we develop a coupling technique by virtue of the dual-branch structure, leading to reduced divergence between the category distributions of the two domains; at the feature level, we introduce an adversarial perturbation strategy to better augment source domain feature, resulting in improved alignment in feature distribution. To the best of our knowledge, our work is the first attempt to alleviate the domain shift issue for WSI data analysis. Extensive experiments on four TCGA datasets have validated the effectiveness of our proposed DETA framework and demonstrated its superior performance in WSI-based survival analysis.

Yuhua Jiang, Qihan Liu, Yiqin Yang et al.

Exploration in sparse reward environments remains a significant challenge in reinforcement learning, particularly in Contextual Markov Decision Processes (CMDPs), where environments differ across episodes. Existing episodic intrinsic motivation methods for CMDPs primarily rely on count-based approaches, which are ineffective in large state spaces, or on similarity-based methods that lack appropriate metrics for state comparison. To address these shortcomings, we propose Episodic Novelty Through Temporal Distance (ETD), a novel approach that introduces temporal distance as a robust metric for state similarity and intrinsic reward computation. By employing contrastive learning, ETD accurately estimates temporal distances and derives intrinsic rewards based on the novelty of states within the current episode. Extensive experiments on various benchmark tasks demonstrate that ETD significantly outperforms state-of-the-art methods, highlighting its effectiveness in enhancing exploration in sparse reward CMDPs.

Yuhua Jiang, Jiawei Huang, Yufeng Yuan et al.

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for enhancing Large Language Models (LLMs) on complex reasoning tasks. However, existing methods suffer from an exploration dilemma: the sharply peaked initial policies of pre-trained LLMs confine standard RL algorithms to a narrow set of solutions, boosting single-solution accuracy (pass@1) but suppressing solution diversity and multi-solution performance (pass@k). As a result, RLVR often distills existing capabilities rather than discovering new reasoning strategies. To overcome this, we introduce a Risk-Sensitive Reinforcement Learning framework. Our approach employs a risk-seeking objective that interpolates between mean and maximum rewards, leading to a novel algorithm, Risk-Sensitive GRPO (RS-GRPO), which drives deeper exploration by amplifying learning from challenging prompts. Remarkably, RS-GRPO is simple to implement, requiring only minor code modifications. On six mathematical reasoning benchmarks and with five different LLMs, RS-GRPO consistently improves pass@k performance while maintaining or enhancing pass@1 accuracy.

Yiqin Yang, Quanwei Wang, Chenghao Li et al.

Offline reinforcement learning (RL) represents a significant shift in RL research, allowing agents to learn from pre-collected datasets without further interaction with the environment. A key, yet underexplored, challenge in offline RL is selecting an optimal subset of the offline dataset that enhances both algorithm performance and training efficiency. Reducing dataset size can also reveal the minimal data requirements necessary for solving similar problems. In response to this challenge, we introduce ReDOR (Reduced Datasets for Offline RL), a method that frames dataset selection as a gradient approximation optimization problem. We demonstrate that the widely used actor-critic framework in RL can be reformulated as a submodular optimization objective, enabling efficient subset selection. To achieve this, we adapt orthogonal matching pursuit (OMP), incorporating several novel modifications tailored for offline RL. Our experimental results show that the data subsets identified by ReDOR not only boost algorithm performance but also do so with significantly lower computational complexity.

Xiaoteng Ma, Yiqin Yang, Hao Hu et al.

Offline reinforcement learning (RL) shows promise of applying RL to real-world problems by effectively utilizing previously collected data. Most existing offline RL algorithms use regularization or constraints to suppress extrapolation error for actions outside the dataset. In this paper, we adopt a different framework, which learns the V-function instead of the Q-function to naturally keep the learning procedure within the support of an offline dataset. To enable effective generalization while maintaining proper conservatism in offline learning, we propose Expectile V-Learning (EVL), which smoothly interpolates between the optimal value learning and behavior cloning. Further, we introduce implicit planning along offline trajectories to enhance learned V-values and accelerate convergence. Together, we present a new offline method called Value-based Episodic Memory (VEM). We provide theoretical analysis for the convergence properties of our proposed VEM method, and empirical results in the D4RL benchmark show that our method achieves superior performance in most tasks, particularly in sparse-reward tasks.

Xiaoteng Ma, Xiaohang Tang, Li Xia et al.

Most of reinforcement learning algorithms optimize the discounted criterion which is beneficial to accelerate the convergence and reduce the variance of estimates. Although the discounted criterion is appropriate for certain tasks such as financial related problems, many engineering problems treat future rewards equally and prefer a long-run average criterion. In this paper, we study the reinforcement learning problem with the long-run average criterion. Firstly, we develop a unified trust region theory with discounted and average criteria and derive a novel performance bound within the trust region with the Perturbation Analysis (PA) theory. Secondly, we propose a practical algorithm named Average Policy Optimization (APO), which improves the value estimation with a novel technique named Average Value Constraint. Finally, experiments are conducted in the continuous control environment MuJoCo. In most tasks, APO performs better than the discounted PPO, which demonstrates the effectiveness of our approach. Our work provides a unified framework of the trust region approach including both the discounted and average criteria, which may complement the framework of reinforcement learning beyond the discounted objectives.

Chenghao Li, Tonghan Wang, Chengjie Wu et al.

Recently, deep multi-agent reinforcement learning (MARL) has shown the promise to solve complex cooperative tasks. Its success is partly because of parameter sharing among agents. However, such sharing may lead agents to behave similarly and limit their coordination capacity. In this paper, we aim to introduce diversity in both optimization and representation of shared multi-agent reinforcement learning. Specifically, we propose an information-theoretical regularization to maximize the mutual information between agents' identities and their trajectories, encouraging extensive exploration and diverse individualized behaviors. In representation, we incorporate agent-specific modules in the shared neural network architecture, which are regularized by L1-norm to promote learning sharing among agents while keeping necessary diversity. Empirical results show that our method achieves state-of-the-art performance on Google Research Football and super hard StarCraft II micromanagement tasks.

Xiaoteng Ma, Yiqin Yang, Chenghao Li et al.

Value-based methods of multi-agent reinforcement learning (MARL), especially the value decomposition methods, have been demonstrated on a range of challenging cooperative tasks. However, current methods pay little attention to the interaction between agents, which is essential to teamwork in games or real life. This limits the efficiency of value-based MARL algorithms in the two aspects: collaborative exploration and value function estimation. In this paper, we propose a novel cooperative MARL algorithm named as interactive actor-critic~(IAC), which models the interaction of agents from the perspectives of policy and value function. On the policy side, a multi-agent joint stochastic policy is introduced by adopting a collaborative exploration module, which is trained by maximizing the entropy-regularized expected return. On the value side, we use the shared attention mechanism to estimate the value function of each agent, which takes the impact of the teammates into consideration. At the implementation level, we extend the value decomposition methods to continuous control tasks and evaluate IAC on benchmark tasks including classic control and multi-agent particle environments. Experimental results indicate that our method outperforms the state-of-the-art approaches and achieves better performance in terms of cooperation.

Chenghao Li, Xiaoteng Ma, Chongjie Zhang et al.

The option framework has shown great promise by automatically extracting temporally-extended sub-tasks from a long-horizon task. Methods have been proposed for concurrently learning low-level intra-option policies and high-level option selection policy. However, existing methods typically suffer from two major challenges: ineffective exploration and unstable updates. In this paper, we present a novel and stable off-policy approach that builds on the maximum entropy model to address these challenges. Our approach introduces an information-theoretical intrinsic reward for encouraging the identification of diverse and effective options. Meanwhile, we utilize a probability inference model to simplify the optimization problem as fitting optimal trajectories. Experimental results demonstrate that our approach significantly outperforms prior on-policy and off-policy methods in a range of Mujoco benchmark tasks while still providing benefits for transfer learning. In these tasks, our approach learns a diverse set of options, each of whose state-action space has strong coherence.

Xiaoteng Ma, Junyao Chen, Li Xia et al.

We present Distributional Soft Actor-Critic (DSAC), a distributional reinforcement learning (RL) algorithm that combines the strengths of distributional information of accumulated rewards and entropy-driven exploration from Soft Actor-Critic (SAC) algorithm. DSAC models the randomness in both action and rewards, surpassing baseline performances on various continuous control tasks. Unlike standard approaches that solely maximize expected rewards, we propose a unified framework for risk-sensitive learning, one that optimizes the risk-related objective while balancing entropy to encourage exploration. Extensive experiments demonstrate DSAC's effectiveness in enhancing agent performances for both risk-neutral and risk-sensitive control tasks.