Bo An

Simin Li, Jun Guo, Jingqiao Xiu et al.

This study probes the vulnerabilities of cooperative multi-agent reinforcement learning (c-MARL) under adversarial attacks, a critical determinant of c-MARL's worst-case performance prior to real-world implementation. Current observation-based attacks, constrained by white-box assumptions, overlook c-MARL's complex multi-agent interactions and cooperative objectives, resulting in impractical and limited attack capabilities. To address these shortcomes, we propose Adversarial Minority Influence (AMI), a practical and strong for c-MARL. AMI is a practical black-box attack and can be launched without knowing victim parameters. AMI is also strong by considering the complex multi-agent interaction and the cooperative goal of agents, enabling a single adversarial agent to unilaterally misleads majority victims to form targeted worst-case cooperation. This mirrors minority influence phenomena in social psychology. To achieve maximum deviation in victim policies under complex agent-wise interactions, our unilateral attack aims to characterize and maximize the impact of the adversary on the victims. This is achieved by adapting a unilateral agent-wise relation metric derived from mutual information, thereby mitigating the adverse effects of victim influence on the adversary. To lead the victims into a jointly detrimental scenario, our targeted attack deceives victims into a long-term, cooperatively harmful situation by guiding each victim towards a specific target, determined through a trial-and-error process executed by a reinforcement learning agent. Through AMI, we achieve the first successful attack against real-world robot swarms and effectively fool agents in simulated environments into collectively worst-case scenarios, including Starcraft II and Multi-agent Mujoco. The source code and demonstrations can be found at: https://github.com/DIG-Beihang/AMI.

Shijie Han, Siyuan Li, Bo An et al.

Multi-agent reinforcement learning (MARL) is a prevalent learning paradigm for solving stochastic games. In most MARL studies, agents in a game are defined as teammates or enemies beforehand, and the relationships among the agents remain fixed throughout the game. However, in real-world problems, the agent relationships are commonly unknown in advance or dynamically changing. Many multi-party interactions start off by asking: who is on my team? This question arises whether it is the first day at the stock exchange or the kindergarten. Therefore, training policies for such situations in the face of imperfect information and ambiguous identities is an important problem that needs to be addressed. In this work, we develop a novel identity detection reinforcement learning (IDRL) framework that allows an agent to dynamically infer the identities of nearby agents and select an appropriate policy to accomplish the task. In the IDRL framework, a relation network is constructed to deduce the identities of other agents by observing the behaviors of the agents. A danger network is optimized to estimate the risk of false-positive identifications. Beyond that, we propose an intrinsic reward that balances the need to maximize external rewards and accurate identification. After identifying the cooperation-competition pattern among the agents, IDRL applies one of the off-the-shelf MARL methods to learn the policy. To evaluate the proposed method, we conduct experiments on Red-10 card-shedding game, and the results show that IDRL achieves superior performance over other state-of-the-art MARL methods. Impressively, the relation network has the par performance to identify the identities of agents with top human players; the danger network reasonably avoids the risk of imperfect identification. The code to reproduce all the reported results is available online at https://github.com/MR-BENjie/IDRL.

Hongxin Wei, Renchunzi Xie, Hao Cheng et al.

Detecting out-of-distribution inputs is critical for safe deployment of machine learning models in the real world. However, neural networks are known to suffer from the overconfidence issue, where they produce abnormally high confidence for both in- and out-of-distribution inputs. In this work, we show that this issue can be mitigated through Logit Normalization (LogitNorm) -- a simple fix to the cross-entropy loss -- by enforcing a constant vector norm on the logits in training. Our method is motivated by the analysis that the norm of the logit keeps increasing during training, leading to overconfident output. Our key idea behind LogitNorm is thus to decouple the influence of output's norm during network optimization. Trained with LogitNorm, neural networks produce highly distinguishable confidence scores between in- and out-of-distribution data. Extensive experiments demonstrate the superiority of LogitNorm, reducing the average FPR95 by up to 42.30% on common benchmarks.

Jianxiong Zhang, Bing Guo, Yuming Jiang et al.

Large reasoning models (LRMs) often generate long, seemingly coherent reasoning traces yet still produce incorrect answers, making hallucination detection challenging. Although trajectories contain useful signals, directly using trace text or vanilla hidden states for detection is brittle: traces vary in form and detectors can overfit to superficial patterns rather than answer validity. We introduce Answer-agreement Representation Shaping (ARS), which learns detection-friendly trace-conditioned representations by explicitly encoding answer stability. ARS generates counterfactual answers through small latent interventions, specifically, perturbing the trace-boundary embedding, and labels each perturbation by whether the resulting answer agrees with the original. It then learns representations that bring answer-agreeing states together and separate answer-disagreeing ones, exposing latent instability indicative of hallucination risk. The shaped embeddings are plug-and-play with existing embedding-based detectors and require no human annotations during training. Experiments demonstrate that ARS consistently improves detection and achieves substantial gains over strong baselines. Code is available at: https://github.com/radiolab-ntu/ars_icml2026.

Wentao Zhang, Mingxuan Zhao, Jincheng Gao et al.

The rapid advancement of Large Language Models (LLMs) has led to a surge of financial benchmarks, evolving from static knowledge evaluation toward interactive trading simulations. However, existing frameworks for evaluating real-time trading largely overlook a critical failure mode: the severe behavioral instability of LLMs in sequential decision-making under financial uncertainty. Through extensive experiments, we show that when deployed as trading agents, LLMs exhibit extreme run-to-run variance, generate inconsistent action sequences even under deterministic decoding, and frequently produce irrational action flipping across adjacent time steps. We attribute these behaviors to the stateless autoregressive nature of LLMs, which lack persistent memory of prior actions, together with their sensitivity to continuous-to-discrete action mappings in portfolio allocation tasks. These deficiencies fundamentally undermine the reliability and reproducibility of many existing online and offline trading benchmarks. To address these limitations, we propose AlphaForgeBench, a principled evaluation framework that redefines LLMs as quantitative researchers rather than stochastic trading agents. Instead of producing discrete trading actions, AlphaForgeBench requires models to generate executable alpha factors and compose factor-based trading strategies grounded in financial knowledge. This paradigm decouples reasoning from execution mechanics, enabling deterministic and reproducible evaluation while remaining aligned with real-world quantitative research workflows. Extensive experiments across multiple state-of-the-art LLMs demonstrate that AlphaForgeBench eliminates execution-induced instability and provides a rigorous benchmark for evaluating financial reasoning, strategy formulation, and alpha discovery. Webpage at https://finbrain-lab-hkustgz.github.io/AlphaForgeBench

Hezhe Qiao, Hanghang Tong, Bo An et al.

Graph anomaly detection (GAD), which aims to identify unusual graph instances (nodes, edges, subgraphs, or graphs), has attracted increasing attention in recent years due to its significance in a wide range of applications. Deep learning approaches, graph neural networks (GNNs) in particular, have been emerging as a promising paradigm for GAD, owing to its strong capability in capturing complex structure and/or node attributes in graph data. Considering the large number of methods proposed for GNN-based GAD, it is of paramount importance to summarize the methodologies and findings in the existing GAD studies, so that we can pinpoint effective model designs for tackling open GAD problems. To this end, in this work we aim to present a comprehensive review of deep learning approaches for GAD. Existing GAD surveys are focused on task-specific discussions, making it difficult to understand the technical insights of existing methods and their limitations in addressing some unique challenges in GAD. To fill this gap, we first discuss the problem complexities and their resulting challenges in GAD, and then provide a systematic review of current deep GAD methods from three novel perspectives of methodology, including GNN backbone design, proxy task design for GAD, and graph anomaly measures. To deepen the discussions, we further propose a taxonomy of 13 fine-grained method categories under these three perspectives to provide more in-depth insights into the model designs and their capabilities. To facilitate the experiments and validation, we also summarize a collection of widely-used GAD datasets and empirical comparison. We further discuss multiple open problems to inspire more future high-quality research. A continuously updated repository for datasets, links to the codes of algorithms, and empirical comparison is available at https://github.com/mala-lab/Awesome-Deep-Graph-Anomaly-Detection.

Longtao Zheng, Rundong Wang, Xinrun Wang et al.

Building agents with large language models (LLMs) for computer control is a burgeoning research area, where the agent receives computer states and performs actions to complete complex tasks. Previous computer agents have demonstrated the benefits of in-context learning (ICL); however, their performance is hindered by several issues. First, the limited context length of LLMs and complex computer states restrict the number of exemplars, as a single webpage can consume the entire context. Second, the exemplars in current methods, such as high-level plans and multi-choice questions, cannot represent complete trajectories, leading to suboptimal performance in long-horizon tasks. Third, existing computer agents rely on task-specific exemplars and overlook the similarity among tasks, resulting in poor generalization to novel tasks. To address these challenges, we introduce Synapse, a computer agent featuring three key components: i) state abstraction, which filters out task-irrelevant information from raw states, allowing more exemplars within the limited context, ii) trajectory-as-exemplar prompting, which prompts the LLM with complete trajectories of the abstracted states and actions to improve multi-step decision-making, and iii) exemplar memory, which stores the embeddings of exemplars and retrieves them via similarity search for generalization to novel tasks. We evaluate Synapse on MiniWoB++, a standard task suite, and Mind2Web, a real-world website benchmark. In MiniWoB++, Synapse achieves a 99.2% average success rate (a 10% relative improvement) across 64 tasks using demonstrations from only 48 tasks. Notably, Synapse is the first ICL method to solve the book-flight task in MiniWoB++. Synapse also exhibits a 56% relative improvement in average step success rate over the previous state-of-the-art prompting scheme in Mind2Web.

Hongxin Wei, Huiping Zhuang, Renchunzi Xie et al.

In the presence of noisy labels, designing robust loss functions is critical for securing the generalization performance of deep neural networks. Cross Entropy (CE) loss has been shown to be not robust to noisy labels due to its unboundedness. To alleviate this issue, existing works typically design specialized robust losses with the symmetric condition, which usually lead to the underfitting issue. In this paper, our key idea is to induce a loss bound at the logit level, thus universally enhancing the noise robustness of existing losses. Specifically, we propose logit clipping (LogitClip), which clamps the norm of the logit vector to ensure that it is upper bounded by a constant. In this manner, CE loss equipped with our LogitClip method is effectively bounded, mitigating the overfitting to examples with noisy labels. Moreover, we present theoretical analyses to certify the noise-tolerant ability of LogitClip. Extensive experiments show that LogitClip not only significantly improves the noise robustness of CE loss, but also broadly enhances the generalization performance of popular robust losses.

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.

Zhe Zhao, Haibin Wen, Yingcheng Wu et al.

Scientific discovery demands intelligence, perseverance, and serendipity across vast search spaces. Today, top scientific capabilities remain siloed--one AI system for biological analysis, another for clinical reasoning, mathematical derivation, or materials simulation--and no pre-designed team can anticipate every skill a question will need. Science Earth is a planet-scale scientific runtime in which any capability--a simulation cluster, a wet-lab robot, a proof engine, a single-cell pipeline--can connect to any other, with collaboration structure emerging from the question itself. Its underlying EACN protocol lets capabilities discover one another, negotiate task ownership, and adjudicate across incompatible evidentiary standards without prior knowledge of who will meet whom. This shifts the organizing challenge from workflow design to open-ended connectivity. Two runs validate this under structurally distinct conditions. In a trans-Pacific higher-order Kuramoto synchronization study, agents identified and corrected a closure-ratio assumption in Ott-Antonsen analytic theory that fails outside the Lorentzian limit, within thirty minutes. In an eight-agent single-cell run on the 4.88M-cell Kang 2024 pan-cancer atlas, heterogeneous capabilities coupled over a 64.9-hour window with one structural external instruction, producing three new result layers and anchoring findings against an independent wet-lab study on an adjacent CCR8- TIGIT+ Treg subset. These cases are a first empirical reading, not a benchmark sweep. They show that when AI capabilities are truly connectable and coordination emerges from the problem, scientific reasoning becomes a distributed, self-correcting process--a step towards scaling AI-native discovery to the planet.

Wanqi Xue, Qingpeng Cai, Zhenghai Xue et al.

Current advances in recommender systems have been remarkably successful in optimizing immediate engagement. However, long-term user engagement, a more desirable performance metric, remains difficult to improve. Meanwhile, recent reinforcement learning (RL) algorithms have shown their effectiveness in a variety of long-term goal optimization tasks. For this reason, RL is widely considered as a promising framework for optimizing long-term user engagement in recommendation. Though promising, the application of RL heavily relies on well-designed rewards, but designing rewards related to long-term user engagement is quite difficult. To mitigate the problem, we propose a novel paradigm, recommender systems with human preferences (or Preference-based Recommender systems), which allows RL recommender systems to learn from preferences about users historical behaviors rather than explicitly defined rewards. Such preferences are easily accessible through techniques such as crowdsourcing, as they do not require any expert knowledge. With PrefRec, we can fully exploit the advantages of RL in optimizing long-term goals, while avoiding complex reward engineering. PrefRec uses the preferences to automatically train a reward function in an end-to-end manner. The reward function is then used to generate learning signals to train the recommendation policy. Furthermore, we design an effective optimization method for PrefRec, which uses an additional value function, expectile regression and reward model pre-training to improve the performance. We conduct experiments on a variety of long-term user engagement optimization tasks. The results show that PrefRec significantly outperforms previous state-of-the-art methods in all the tasks.

Wanqi Xue, Bo An, Shuicheng Yan et al.

The complexity of designing reward functions has been a major obstacle to the wide application of deep reinforcement learning (RL) techniques. Describing an agent's desired behaviors and properties can be difficult, even for experts. A new paradigm called reinforcement learning from human preferences (or preference-based RL) has emerged as a promising solution, in which reward functions are learned from human preference labels among behavior trajectories. However, existing methods for preference-based RL are limited by the need for accurate oracle preference labels. This paper addresses this limitation by developing a method for crowd-sourcing preference labels and learning from diverse human preferences. The key idea is to stabilize reward learning through regularization and correction in a latent space. To ensure temporal consistency, a strong constraint is imposed on the reward model that forces its latent space to be close to the prior distribution. Additionally, a confidence-based reward model ensembling method is designed to generate more stable and reliable predictions. The proposed method is tested on a variety of tasks in DMcontrol and Meta-world and has shown consistent and significant improvements over existing preference-based RL algorithms when learning from diverse feedback, paving the way for real-world applications of RL methods.

Zhenghai Xue, Qingpeng Cai, Shuchang Liu et al.

In many real-world scenarios, Reinforcement Learning (RL) algorithms are trained on data with dynamics shift, i.e., with different underlying environment dynamics. A majority of current methods address such issue by training context encoders to identify environment parameters. Data with dynamics shift are separated according to their environment parameters to train the corresponding policy. However, these methods can be sample inefficient as data are used \textit{ad hoc}, and policies trained for one dynamics cannot benefit from data collected in all other environments with different dynamics. In this paper, we find that in many environments with similar structures and different dynamics, optimal policies have similar stationary state distributions. We exploit such property and learn the stationary state distribution from data with dynamics shift for efficient data reuse. Such distribution is used to regularize the policy trained in a new environment, leading to the SRPO (\textbf{S}tate \textbf{R}egularized \textbf{P}olicy \textbf{O}ptimization) algorithm. To conduct theoretical analyses, the intuition of similar environment structures is characterized by the notion of homomorphous MDPs. We then demonstrate a lower-bound performance guarantee on policies regularized by the stationary state distribution. In practice, SRPO can be an add-on module to context-based algorithms in both online and offline RL settings. Experimental results show that SRPO can make several context-based algorithms far more data efficient and significantly improve their overall performance.

Hongxin Wei, Lue Tao, Renchunzi Xie et al.

Deep neural networks usually perform poorly when the training dataset suffers from extreme class imbalance. Recent studies found that directly training with out-of-distribution data (i.e., open-set samples) in a semi-supervised manner would harm the generalization performance. In this work, we theoretically show that out-of-distribution data can still be leveraged to augment the minority classes from a Bayesian perspective. Based on this motivation, we propose a novel method called Open-sampling, which utilizes open-set noisy labels to re-balance the class priors of the training dataset. For each open-set instance, the label is sampled from our pre-defined distribution that is complementary to the distribution of original class priors. We empirically show that Open-sampling not only re-balances the class priors but also encourages the neural network to learn separable representations. Extensive experiments demonstrate that our proposed method significantly outperforms existing data re-balancing methods and can boost the performance of existing state-of-the-art methods.

Haochong Xia, Shuo Sun, Xinrun Wang et al.

Financial simulators play an important role in enhancing forecasting accuracy, managing risks, and fostering strategic financial decision-making. Despite the development of financial market simulation methodologies, existing frameworks often struggle with adapting to specialized simulation context. We pinpoint the challenges as i) current financial datasets do not contain context labels; ii) current techniques are not designed to generate financial data with context as control, which demands greater precision compared to other modalities; iii) the inherent difficulties in generating context-aligned, high-fidelity data given the non-stationary, noisy nature of financial data. To address these challenges, our contributions are: i) we proposed the Contextual Market Dataset with market dynamics, stock ticker, and history state as context, leveraging a market dynamics modeling method that combines linear regression and Dynamic Time Warping clustering to extract market dynamics; ii) we present Market-GAN, a novel architecture incorporating a Generative Adversarial Networks (GAN) for the controllable generation with context, an autoencoder for learning low-dimension features, and supervisors for knowledge transfer; iii) we introduce a two-stage training scheme to ensure that Market-GAN captures the intrinsic market distribution with multiple objectives. In the pertaining stage, with the use of the autoencoder and supervisors, we prepare the generator with a better initialization for the adversarial training stage. We propose a set of holistic evaluation metrics that consider alignment, fidelity, data usability on downstream tasks, and market facts. We evaluate Market-GAN with the Dow Jones Industrial Average data from 2000 to 2023 and showcase superior performance in comparison to 4 state-of-the-art time-series generative models.

Shuo He, Lang Feng, Qi Wei et al.

Group-based reinforcement learning (RL), such as GRPO, has advanced the capabilities of large language models on long-horizon agentic tasks. To enable more fine-grained policy updates, recent research has increasingly shifted toward stepwise group-based policy optimization, which treats each step in a rollout trajectory independently while using a memory module to retain historical context. However, we find a key issue in estimating stepwise relative advantages, namely context inconsistency, where steps within the same group may differ in their historical contexts. Empirically, we reveal that this issue can lead to severely biased advantage estimation, thereby degrading policy optimization significantly. To address the issue, in this paper, we propose Hierarchy-of-Groups Policy Optimization (HGPO) for long-horizon agentic tasks. Specifically, within a group of rollout trajectories, HGPO assigns each step to multiple hierarchical groups according to the consistency of historical contexts. Then, for each step, HGPO computes distinct advantages within each group and aggregates them with an adaptive weighting scheme. In this way, HGPO can achieve a favorable bias-variance trade-off in stepwise advantage estimation, without extra models or rollouts. Evaluations on two challenging agentic tasks, ALFWorld and WebShop with Qwen2.5-1.5B-Instruct and Qwen2.5-7B-Instruct, show that HGPO significantly outperforms existing agentic RL methods under the same computational constraints. Code is available at https://github.com/langfengQ/verl-agent/tree/master/recipe/hgpo.

Shuo Sun, Molei Qin, Xinrun Wang et al.

The financial markets, which involve more than $90 trillion market capitals, attract the attention of innumerable investors around the world. Recently, reinforcement learning in financial markets (FinRL) has emerged as a promising direction to train agents for making profitable investment decisions. However, the evaluation of most FinRL methods only focuses on profit-related measures and ignores many critical axes, which are far from satisfactory for financial practitioners to deploy these methods into real-world financial markets. Therefore, we introduce PRUDEX-Compass, which has 6 axes, i.e., Profitability, Risk-control, Universality, Diversity, rEliability, and eXplainability, with a total of 17 measures for a systematic evaluation. Specifically, i) we propose AlphaMix+ as a strong FinRL baseline, which leverages mixture-of-experts (MoE) and risk-sensitive approaches to make diversified risk-aware investment decisions, ii) we evaluate 8 FinRL methods in 4 long-term real-world datasets of influential financial markets to demonstrate the usage of our PRUDEX-Compass, iii) PRUDEX-Compass together with 4 real-world datasets, standard implementation of 8 FinRL methods and a portfolio management environment is released as public resources to facilitate the design and comparison of new FinRL methods. We hope that PRUDEX-Compass can not only shed light on future FinRL research to prevent untrustworthy results from stagnating FinRL into successful industry deployment but also provide a new challenging algorithm evaluation scenario for the reinforcement learning (RL) community.

Wanqi Xue, Qingpeng Cai, Ruohan Zhan et al.

Long-term engagement is preferred over immediate engagement in sequential recommendation as it directly affects product operational metrics such as daily active users (DAUs) and dwell time. Meanwhile, reinforcement learning (RL) is widely regarded as a promising framework for optimizing long-term engagement in sequential recommendation. However, due to expensive online interactions, it is very difficult for RL algorithms to perform state-action value estimation, exploration and feature extraction when optimizing long-term engagement. In this paper, we propose ResAct which seeks a policy that is close to, but better than, the online-serving policy. In this way, we can collect sufficient data near the learned policy so that state-action values can be properly estimated, and there is no need to perform online exploration. ResAct optimizes the policy by first reconstructing the online behaviors and then improving it via a Residual Actor. To extract long-term information, ResAct utilizes two information-theoretical regularizers to confirm the expressiveness and conciseness of features. We conduct experiments on a benchmark dataset and a large-scale industrial dataset which consists of tens of millions of recommendation requests. Experimental results show that our method significantly outperforms the state-of-the-art baselines in various long-term engagement optimization tasks.

Hui Niu, Siyuan Li, Jiahao Zheng et al.

Market making (MM) has attracted significant attention in financial trading owing to its essential function in ensuring market liquidity. With strong capabilities in sequential decision-making, Reinforcement Learning (RL) technology has achieved remarkable success in quantitative trading. Nonetheless, most existing RL-based MM methods focus on optimizing single-price level strategies which fail at frequent order cancellations and loss of queue priority. Strategies involving multiple price levels align better with actual trading scenarios. However, given the complexity that multi-price level strategies involves a comprehensive trading action space, the challenge of effectively training profitable RL agents for MM persists. Inspired by the efficient workflow of professional human market makers, we propose Imitative Market Maker (IMM), a novel RL framework leveraging both knowledge from suboptimal signal-based experts and direct policy interactions to develop multi-price level MM strategies efficiently. The framework start with introducing effective state and action representations adept at encoding information about multi-price level orders. Furthermore, IMM integrates a representation learning unit capable of capturing both short- and long-term market trends to mitigate adverse selection risk. Subsequently, IMM formulates an expert strategy based on signals and trains the agent through the integration of RL and imitation learning techniques, leading to efficient learning. Extensive experimental results on four real-world market datasets demonstrate that IMM outperforms current RL-based market making strategies in terms of several financial criteria. The findings of the ablation study substantiate the effectiveness of the model components.

Xin Cheng, Deng-Bao Wang, Lei Feng et al.

Partial-label learning is a popular weakly supervised learning setting that allows each training example to be annotated with a set of candidate labels. Previous studies on partial-label learning only focused on the classification setting where candidate labels are all discrete, which cannot handle continuous labels with real values. In this paper, we provide the first attempt to investigate partial-label regression, where each training example is annotated with a set of real-valued candidate labels. To solve this problem, we first propose a simple baseline method that takes the average loss incurred by candidate labels as the predictive loss. The drawback of this method lies in that the loss incurred by the true label may be overwhelmed by other false labels. To overcome this drawback, we propose an identification method that takes the least loss incurred by candidate labels as the predictive loss. We further improve it by proposing a progressive identification method to differentiate candidate labels using progressively updated weights for incurred losses. We prove that the latter two methods are model-consistent and provide convergence analyses. Our proposed methods are theoretically grounded and can be compatible with any models, optimizers, and losses. Experiments validate the effectiveness of our proposed methods.

Shuo Sun, Rundong Wang, Bo An

Quantitative investment is a fundamental financial task that highly relies on accurate stock prediction and profitable investment decision making. Despite recent advances in deep learning (DL) have shown stellar performance on capturing trading opportunities in the stochastic stock market, we observe that the performance of existing DL methods is sensitive to random seeds and network initialization. To design more profitable DL methods, we analyze this phenomenon and find two major limitations of existing works. First, there is a noticeable gap between accurate financial predictions and profitable investment strategies. Second, investment decisions are made based on only one individual predictor without consideration of model uncertainty, which is inconsistent with the workflow in real-world trading firms. To tackle these two limitations, we first reformulate quantitative investment as a multi-task learning problem. Later on, we propose AlphaMix, a novel two-stage mixture-of-experts (MoE) framework for quantitative investment to mimic the efficient bottom-up trading strategy design workflow of successful trading firms. In Stage one, multiple independent trading experts are jointly optimized with an individual uncertainty-aware loss function. In Stage two, we train neural routers (corresponding to the role of a portfolio manager) to dynamically deploy these experts on an as-needed basis. AlphaMix is also a universal framework that is applicable to various backbone network architectures with consistent performance gains. Through extensive experiments on long-term real-world data spanning over five years on two of the most influential financial markets (US and China), we demonstrate that AlphaMix significantly outperforms many state-of-the-art baselines in terms of four financial criteria.

Rundong Wang, Longtao Zheng, Wei Qiu et al.

Recent advances in multi-agent reinforcement learning (MARL) allow agents to coordinate their behaviors in complex environments. However, common MARL algorithms still suffer from scalability and sparse reward issues. One promising approach to resolving them is automatic curriculum learning (ACL). ACL involves a student (curriculum learner) training on tasks of increasing difficulty controlled by a teacher (curriculum generator). Despite its success, ACL's applicability is limited by (1) the lack of a general student framework for dealing with the varying number of agents across tasks and the sparse reward problem, and (2) the non-stationarity of the teacher's task due to ever-changing student strategies. As a remedy for ACL, we introduce a novel automatic curriculum learning framework, Skilled Population Curriculum (SPC), which adapts curriculum learning to multi-agent coordination. Specifically, we endow the student with population-invariant communication and a hierarchical skill set, allowing it to learn cooperation and behavior skills from distinct tasks with varying numbers of agents. In addition, we model the teacher as a contextual bandit conditioned by student policies, enabling a team of agents to change its size while still retaining previously acquired skills. We also analyze the inherent non-stationarity of this multi-agent automatic curriculum teaching problem and provide a corresponding regret bound. Empirical results show that our method improves the performance, scalability and sample efficiency in several MARL environments.

Yuzhou Cao, Hussein Mozannar, Lei Feng et al.

Enabling machine learning classifiers to defer their decision to a downstream expert when the expert is more accurate will ensure improved safety and performance. This objective can be achieved with the learning-to-defer framework which aims to jointly learn how to classify and how to defer to the expert. In recent studies, it has been theoretically shown that popular estimators for learning to defer parameterized with softmax provide unbounded estimates for the likelihood of deferring which makes them uncalibrated. However, it remains unknown whether this is due to the widely used softmax parameterization and if we can find a softmax-based estimator that is both statistically consistent and possesses a valid probability estimator. In this work, we first show that the cause of the miscalibrated and unbounded estimator in prior literature is due to the symmetric nature of the surrogate losses used and not due to softmax. We then propose a novel statistically consistent asymmetric softmax-based surrogate loss that can produce valid estimates without the issue of unboundedness. We further analyze the non-asymptotic properties of our method and empirically validate its performance and calibration on benchmark datasets.

Zhenglin Wan, Jingxuan Wu, Xingrui Yu et al.

Learning from demonstrations in embodied control is often cast as behavioral cloning, and recent diffusion or flow-matching policies improve this paradigm by modeling multi-modal expert actions. Yet these methods remain offline supervised learners: the policy is trained only on expert states and receives no corrective signal on the states it actually visits. On-policy distillation (OPD) offers a natural remedy, but standard OPD assumes a strong fixed teacher, which is unavailable in demonstration-only control. We propose \textbf{FA-OPD}, an \emph{adversarial dual on-policy distillation} method in which a Flow Matching (FM) teacher is learned from demonstrations and co-trained with a lightweight MLP student. The teacher provides two complementary signals on student rollouts. The reward channel learns an expert-likeness objective over state-action pairs and drives online exploration through long-horizon policy optimization. The action channel supplies dense local targets at student-visited states, stabilizing exploitation. FA-OPD couples them so that reward distillation enables generalization beyond point-wise demonstrations, while action distillation keeps exploration anchored near expert-like behavior. Across six robot navigation, manipulation, and locomotion benchmarks, FA-OPD beats strong baselines and shows much stronger robustness under noisy or limited demonstrations.

Xin Cheng, Shuo He, Lang Feng et al.

Group-based reinforcement learning (RL) methods have achieved remarkable success in improving the performance of large language models (LLMs) and have been rapidly extended to agentic tasks. However, their credit assignment relies heavily on coarse-grained trajectory-level attribution according to final outcomes, making it difficult to capture the contribution of individual steps, such as valuable steps obscured within failed trajectories. To uncover latent information and enable more faithful step-level credit assignment, we propose Graph-based Group Policy Optimization (GraphGPO), which first aggregates all rollout trajectories into a unified state-transition graph and then estimates the distance from each state to the task goal using the global information encoded in the graph. Finally, GraphGPO assigns credit to each edge by estimating a graph-based advantage, based on how much the transition reduces the distance to the task goal. In this way, GraphGPO significantly improves training efficiency and achieves state-of-the-art performance across a range of challenging benchmarks.

Wei Qiu, Xiao Ma, Bo An et al.

Despite the recent advancement in multi-agent reinforcement learning (MARL), the MARL agents easily overfit the training environment and perform poorly in the evaluation scenarios where other agents behave differently. Obtaining generalizable policies for MARL agents is thus necessary but challenging mainly due to complex multi-agent interactions. In this work, we model the problem with Markov Games and propose a simple yet effective method, ranked policy memory (RPM), to collect diverse multi-agent trajectories for training MARL policies with good generalizability. The main idea of RPM is to maintain a look-up memory of policies. In particular, we try to acquire various levels of behaviors by saving policies via ranking the training episode return, i.e., the episode return of agents in the training environment; when an episode starts, the learning agent can then choose a policy from the RPM as the behavior policy. This innovative self-play training framework leverages agents' past policies and guarantees the diversity of multi-agent interaction in the training data. We implement RPM on top of MARL algorithms and conduct extensive experiments on Melting Pot. It has been demonstrated that RPM enables MARL agents to interact with unseen agents in multi-agent generalization evaluation scenarios and complete given tasks, and it significantly boosts the performance up to 402% on average.

Wei Qiu, Weixun Wang, Rundong Wang et al.

We investigate model-free multi-agent reinforcement learning (MARL) in environments where off-beat actions are prevalent, i.e., all actions have pre-set execution durations. During execution durations, the environment changes are influenced by, but not synchronised with, action execution. Such a setting is ubiquitous in many real-world problems. However, most MARL methods assume actions are executed immediately after inference, which is often unrealistic and can lead to catastrophic failure for multi-agent coordination with off-beat actions. In order to fill this gap, we develop an algorithmic framework for MARL with off-beat actions. We then propose a novel episodic memory, LeGEM, for model-free MARL algorithms. LeGEM builds agents' episodic memories by utilizing agents' individual experiences. It boosts multi-agent learning by addressing the challenging temporal credit assignment problem raised by the off-beat actions via our novel reward redistribution scheme, alleviating the issue of non-Markovian reward. We evaluate LeGEM on various multi-agent scenarios with off-beat actions, including Stag-Hunter Game, Quarry Game, Afforestation Game, and StarCraft II micromanagement tasks. Empirical results show that LeGEM significantly boosts multi-agent coordination and achieves leading performance and improved sample efficiency.

Xin Cheng, Yuzhou Cao, Ximing Li et al.

This paper investigates an interesting weakly supervised regression setting called regression with interval targets (RIT). Although some of the previous methods on relevant regression settings can be adapted to RIT, they are not statistically consistent, and thus their empirical performance is not guaranteed. In this paper, we provide a thorough study on RIT. First, we proposed a novel statistical model to describe the data generation process for RIT and demonstrate its validity. Second, we analyze a simple selection method for RIT, which selects a particular value in the interval as the target value to train the model. Third, we propose a statistically consistent limiting method for RIT to train the model by limiting the predictions to the interval. We further derive an estimation error bound for our limiting method. Finally, extensive experiments on various datasets demonstrate the effectiveness of our proposed method.

Wentao Zhang, Zhe Zhao, Haibin Wen et al.

Recent advances in LLM based agent systems have shown promise in tackling complex, long horizon tasks. However, existing agent protocols (e.g., A2A and MCP) under specify cross entity lifecycle and context management, version tracking, and evolution safe update interfaces, which encourages monolithic compositions and brittle glue code. We introduce Autogenesis Protocol (AGP), a self evolution protocol that decouples what evolves from how evolution occurs. Its Resource Substrate Protocol Layer (RSPL) models prompts, agents, tools, environments, and memory as protocol registered resources with explicit state, lifecycle, and versioned interfaces. Its Self Evolution Protocol Layer (SEPL) specifies a closed loop operator interface for proposing, assessing, and committing improvements with auditable lineage and rollback. Building on AGP, we present Autogenesis System (AGS), a self-evolving multi-agent system that dynamically instantiates, retrieves, and refines protocol-registered resources during execution. We evaluate AGS on multiple challenging benchmarks that require long horizon planning and tool use across heterogeneous resources. The results demonstrate consistent improvements over strong baselines, supporting the effectiveness of agent resource management and closed loop self evolution. The code is available at https://github.com/DVampire/Autogenesis.

Renchunzi Xie, Hongxin Wei, Lei Feng et al.

Estimating the generalization performance is practically challenging on out-of-distribution (OOD) data without ground-truth labels. While previous methods emphasize the connection between distribution difference and OOD accuracy, we show that a large domain gap not necessarily leads to a low test accuracy. In this paper, we investigate this problem from the perspective of feature separability empirically and theoretically. Specifically, we propose a dataset-level score based upon feature dispersion to estimate the test accuracy under distribution shift. Our method is inspired by desirable properties of features in representation learning: high inter-class dispersion and high intra-class compactness. Our analysis shows that inter-class dispersion is strongly correlated with the model accuracy, while intra-class compactness does not reflect the generalization performance on OOD data. Extensive experiments demonstrate the superiority of our method in both prediction performance and computational efficiency.

Penghui Yang, Zhonghan Zhang, Yue Li et al.

Density functional theory (DFT) serves as the basis for computational discovery in materials science and chemistry, yet each calculation demands extensive human effort: adjusting algorithms when convergence stalls, revising plans when unexpected physics emerges, and inserting steps as intermediate results reshape the problem. Existing LLM-based agents automate only the initial planning stage, producing a full execution plan upfront and leaving all subsequent adaptation to hand-crafted rules. As a result, these workflows remain fragile, do not generalize well beyond pre-planned scenarios, and often require expert intervention when failures or unexpected intermediate results require changes to the calculation path. Here, we introduce AutoDFT, a closed-loop multi-agent framework that embeds LLM reasoning into every stage of the DFT lifecycle, where a strategic planner produces a skeletal plan of step objectives; a step planner generates numerical parameters just in time from preceding results; and a monitor-recover-reflect cycle diagnoses failures, repairs them, and revises the plan when the evidence justifies it. We demonstrate both breadth and depth: breadth on VASPBench, a purpose-built benchmark spanning 34 tasks and 9 DFT calculation types, where AutoDFT achieves 94.1% task-level success with GPT-5.2; and depth on established materials databases, where AutoDFT produces quantitatively reliable property predictions across electronic, magnetic, and energetic properties. By closing the loop between planning and execution, AutoDFT enables experimentalists without deep computational expertise to obtain reliable first-principles results.

Wentao Zhang, Yilei Zhao, Shuo Sun et al.

Portfolio management (PM) is a fundamental financial trading task, which explores the optimal periodical reallocation of capitals into different stocks to pursue long-term profits. Reinforcement learning (RL) has recently shown its potential to train profitable agents for PM through interacting with financial markets. However, existing work mostly focuses on fixed stock pools, which is inconsistent with investors' practical demand. Specifically, the target stock pool of different investors varies dramatically due to their discrepancy on market states and individual investors may temporally adjust stocks they desire to trade (e.g., adding one popular stocks), which lead to customizable stock pools (CSPs). Existing RL methods require to retrain RL agents even with a tiny change of the stock pool, which leads to high computational cost and unstable performance. To tackle this challenge, we propose EarnMore, a rEinforcement leARNing framework with Maskable stOck REpresentation to handle PM with CSPs through one-shot training in a global stock pool (GSP). Specifically, we first introduce a mechanism to mask out the representation of the stocks outside the target pool. Second, we learn meaningful stock representations through a self-supervised masking and reconstruction process. Third, a re-weighting mechanism is designed to make the portfolio concentrate on favorable stocks and neglect the stocks outside the target pool. Through extensive experiments on 8 subset stock pools of the US stock market, we demonstrate that EarnMore significantly outperforms 14 state-of-the-art baselines in terms of 6 popular financial metrics with over 40% improvement on profit.

Xin Cheng, Yuzhou Cao, Haobo Wang et al.

Learning with rejection is an important framework that can refrain from making predictions to avoid critical mispredictions by balancing between prediction and rejection. Previous studies on cost-based rejection only focused on the classification setting, which cannot handle the continuous and infinite target space in the regression setting. In this paper, we investigate a novel regression problem called regression with cost-based rejection, where the model can reject to make predictions on some examples given certain rejection costs. To solve this problem, we first formulate the expected risk for this problem and then derive the Bayes optimal solution, which shows that the optimal model should reject to make predictions on the examples whose variance is larger than the rejection cost when the mean squared error is used as the evaluation metric. Furthermore, we propose to train the model by a surrogate loss function that considers rejection as binary classification and we provide conditions for the model consistency, which implies that the Bayes optimal solution can be recovered by our proposed surrogate loss. Extensive experiments demonstrate the effectiveness of our proposed method.

Pengdeng Li, Xinrun Wang, Shuxin Li et al.

In this work, we attempt to bridge the two fields of finite-agent and infinite-agent games, by studying how the optimal policies of agents evolve with the number of agents (population size) in mean-field games, an agent-centric perspective in contrast to the existing works focusing typically on the convergence of the empirical distribution of the population. To this end, the premise is to obtain the optimal policies of a set of finite-agent games with different population sizes. However, either deriving the closed-form solution for each game is theoretically intractable, training a distinct policy for each game is computationally intensive, or directly applying the policy trained in a game to other games is sub-optimal. We address these challenges through the Population-size-Aware Policy Optimization (PAPO). Our contributions are three-fold. First, to efficiently generate efficient policies for games with different population sizes, we propose PAPO, which unifies two natural options (augmentation and hypernetwork) and achieves significantly better performance. PAPO consists of three components: i) the population-size encoding which transforms the original value of population size to an equivalent encoding to avoid training collapse, ii) a hypernetwork to generate a distinct policy for each game conditioned on the population size, and iii) the population size as an additional input to the generated policy. Next, we construct a multi-task-based training procedure to efficiently train the neural networks of PAPO by sampling data from multiple games with different population sizes. Finally, extensive experiments on multiple environments show the significant superiority of PAPO over baselines, and the analysis of the evolution of the generated policies further deepens our understanding of the two fields of finite-agent and infinite-agent games.

Shuxin Li, Xinrun Wang, Youzhi Zhang et al.

Offline reinforcement learning (offline RL) is an emerging field that has recently begun gaining attention across various application domains due to its ability to learn strategies from earlier collected datasets. Offline RL proved very successful, paving a path to solving previously intractable real-world problems, and we aim to generalize this paradigm to a multiplayer-game setting. To this end, we introduce a problem of offline equilibrium finding (OEF) and construct multiple types of datasets across a wide range of games using several established methods. To solve the OEF problem, we design a model-based framework that can directly apply any online equilibrium finding algorithm to the OEF setting while making minimal changes. The three most prominent contemporary online equilibrium finding algorithms are adapted to the context of OEF, creating three model-based variants: OEF-PSRO and OEF-CFR, which generalize the widely-used algorithms PSRO and Deep CFR to compute Nash equilibria (NEs), and OEF-JPSRO, which generalizes the JPSRO to calculate (Coarse) Correlated equilibria ((C)CEs). We also combine the behavior cloning policy with the model-based policy to further improve the performance and provide a theoretical guarantee of the solution quality. Extensive experimental results demonstrate the superiority of our approach over offline RL algorithms and the importance of using model-based methods for OEF problems. We hope our work will contribute to advancing research in large-scale equilibrium finding.

Jujie He, Jiacai Liu, Chris Yuhao Liu et al.

The success of DeepSeek-R1 underscores the significant role of reinforcement learning (RL) in enhancing the reasoning capabilities of large language models (LLMs). In this work, we present Skywork-OR1, an effective and scalable RL implementation for long Chain-of-Thought (CoT) models. Building on the DeepSeek-R1-Distill model series, our RL approach achieves notable performance gains, increasing average accuracy across AIME24, AIME25, and LiveCodeBench from 57.8% to 72.8% (+15.0%) for the 32B model and from 43.6% to 57.5% (+13.9%) for the 7B model. Our Skywork-OR1-32B model surpasses both DeepSeek-R1 and Qwen3-32B on the AIME24 and AIME25 benchmarks, while achieving comparable results on LiveCodeBench. The Skywork-OR1-7B and Skywork-OR1-Math-7B models demonstrate competitive reasoning capabilities among models of similar size. We perform comprehensive ablation studies on the core components of our training pipeline to validate their effectiveness. Additionally, we thoroughly investigate the phenomenon of entropy collapse, identify key factors affecting entropy dynamics, and demonstrate that mitigating premature entropy collapse is critical for improved test performance. To support community research, we fully open-source our model weights, training code, and training datasets.

Zhenghai Xue, Qingpeng Cai, Bin Yang et al.

The field of Reinforcement Learning (RL) has garnered increasing attention for its ability of optimizing user retention in recommender systems. A primary obstacle in this optimization process is the environment non-stationarity stemming from the continual and complex evolution of user behavior patterns over time, such as variations in interaction rates and retention propensities. These changes pose significant challenges to existing RL algorithms for recommendations, leading to issues with dynamics and reward distribution shifts. This paper introduces a novel approach called \textbf{A}daptive \textbf{U}ser \textbf{R}etention \textbf{O}ptimization (AURO) to address this challenge. To navigate the recommendation policy in non-stationary environments, AURO introduces an state abstraction module in the policy network. The module is trained with a new value-based loss function, aligning its output with the estimated performance of the current policy. As the policy performance of RL is sensitive to environment drifts, the loss function enables the state abstraction to be reflective of environment changes and notify the recommendation policy to adapt accordingly. Additionally, the non-stationarity of the environment introduces the problem of implicit cold start, where the recommendation policy continuously interacts with users displaying novel behavior patterns. AURO encourages exploration guarded by performance-based rejection sampling to maintain a stable recommendation quality in the cost-sensitive online environment. Extensive empirical analysis are conducted in a user retention simulator, the MovieLens dataset, and a live short-video recommendation platform, demonstrating AURO's superior performance against all evaluated baseline algorithms.

Yewen Li, Chaojie Wang, Xiaobo Xia et al.

Unsupervised out-of-distribution (U-OOD) detection is to identify OOD data samples with a detector trained solely on unlabeled in-distribution (ID) data. The likelihood function estimated by a deep generative model (DGM) could be a natural detector, but its performance is limited in some popular "hard" benchmarks, such as FashionMNIST (ID) vs. MNIST (OOD). Recent studies have developed various detectors based on DGMs to move beyond likelihood. However, despite their success on "hard" benchmarks, most of them struggle to consistently surpass or match the performance of likelihood on some "non-hard" cases, such as SVHN (ID) vs. CIFAR10 (OOD) where likelihood could be a nearly perfect detector. Therefore, we appeal for more attention to incremental effectiveness on likelihood, i.e., whether a method could always surpass or at least match the performance of likelihood in U-OOD detection. We first investigate the likelihood of variational DGMs and find its detection performance could be improved in two directions: i) alleviating latent distribution mismatch, and ii) calibrating the dataset entropy-mutual integration. Then, we apply two techniques for each direction, specifically post-hoc prior and dataset entropy-mutual calibration. The final method, named Resultant, combines these two directions for better incremental effectiveness compared to either technique alone. Experimental results demonstrate that the Resultant could be a new state-of-the-art U-OOD detector while maintaining incremental effectiveness on likelihood in a wide range of tasks.

Yanchen Deng, Chendong Zhao, Yixuan Li et al.

The discovery of advanced metallic alloys is hindered by vast composition spaces, competing property objectives, and real-world constraints on manufacturability. Here we introduce MATAI, a generalist machine learning framework for property prediction and inverse design of as-cast alloys. MATAI integrates a curated alloy database, deep neural network-based property predictors, a constraint-aware optimization engine, and an iterative AI-experiment feedback loop. The framework estimates key mechanical propertie, sincluding density, yield strength, ultimate tensile strength, and elongation, directly from composition, using multi-task learning and physics-informed inductive biases. Alloy design is framed as a constrained optimization problem and solved using a bi-level approach that combines local search with symbolic constraint programming. We demonstrate MATAI's capabilities on the Ti-based alloy system, a canonical class of lightweight structural materials, where it rapidly identifies candidates that simultaneously achieve lower density (<4.45 g/cm3), higher strength (>1000 MPa) and appreciable ductility (>5%) through only seven iterations. Experimental validation confirms that MATAI-designed alloys outperform commercial references such as TC4, highlighting the framework's potential to accelerate the discovery of lightweight, high-performance materials under real-world design constraints.

Yanchen Deng, Shufeng Kong, Caihua Liu et al.

Belief Propagation (BP) is an important message-passing algorithm for various reasoning tasks over graphical models, including solving the Constraint Optimization Problems (COPs). It has been shown that BP can achieve state-of-the-art performance on various benchmarks by mixing old and new messages before sending the new one, i.e., damping. However, existing methods of tuning a static damping factor for BP not only are laborious but also harm their performance. Moreover, existing BP algorithms treat each variable node's neighbors equally when composing a new message, which also limits their exploration ability. To address these issues, we seamlessly integrate BP, Gated Recurrent Units (GRUs), and Graph Attention Networks (GATs) within the message-passing framework to reason about dynamic weights and damping factors for composing new BP messages. Our model, Deep Attentive Belief Propagation (DABP), takes the factor graph and the BP messages in each iteration as the input and infers the optimal weights and damping factors through GRUs and GATs, followed by a multi-head attention layer. Furthermore, unlike existing neural-based BP variants, we propose a novel self-supervised learning algorithm for DABP with a smoothed solution cost, which does not require expensive training labels and also avoids the common out-of-distribution issue through efficient online learning. Extensive experiments show that our model significantly outperforms state-of-the-art baselines.

Safa Messaoud, Billel Mokeddem, Zhenghai Xue et al.

Learning expressive stochastic policies instead of deterministic ones has been proposed to achieve better stability, sample complexity, and robustness. Notably, in Maximum Entropy Reinforcement Learning (MaxEnt RL), the policy is modeled as an expressive Energy-Based Model (EBM) over the Q-values. However, this formulation requires the estimation of the entropy of such EBMs, which is an open problem. To address this, previous MaxEnt RL methods either implicitly estimate the entropy, resulting in high computational complexity and variance (SQL), or follow a variational inference procedure that fits simplified actor distributions (e.g., Gaussian) for tractability (SAC). We propose Stein Soft Actor-Critic (S$^2$AC), a MaxEnt RL algorithm that learns expressive policies without compromising efficiency. Specifically, S$^2$AC uses parameterized Stein Variational Gradient Descent (SVGD) as the underlying policy. We derive a closed-form expression of the entropy of such policies. Our formula is computationally efficient and only depends on first-order derivatives and vector products. Empirical results show that S$^2$AC yields more optimal solutions to the MaxEnt objective than SQL and SAC in the multi-goal environment, and outperforms SAC and SQL on the MuJoCo benchmark. Our code is available at: https://github.com/SafaMessaoud/S2AC-Energy-Based-RL-with-Stein-Soft-Actor-Critic

Haochong Xia, Yao Long Teng, Regan Tan et al.

In quantitative finance, the gap between training and real-world performance-driven by concept drift and distributional non-stationarity-remains a critical obstacle for building reliable data-driven systems. Models trained on static historical data often overfit, resulting in poor generalization in dynamic markets. The mantra "History Is Not Enough" underscores the need for adaptive data generation that learns to evolve with the market rather than relying solely on past observations. We present a drift-aware dataflow system that integrates machine learning-based adaptive control into the data curation process. The system couples a parameterized data manipulation module comprising single-stock transformations, multi-stock mix-ups, and curation operations, with an adaptive planner-scheduler that employs gradient-based bi-level optimization to control the system. This design unifies data augmentation, curriculum learning, and data workflow management under a single differentiable framework, enabling provenance-aware replay and continuous data quality monitoring. Extensive experiments on forecasting and reinforcement learning trading tasks demonstrate that our framework enhances model robustness and improves risk-adjusted returns. The system provides a generalizable approach to adaptive data management and learning-guided workflow automation for financial data.

Molei Qin, Xinyu Cai, Yewen Li et al.

Futures are contracts obligating the exchange of an asset at a predetermined date and price, notable for their high leverage and liquidity and, therefore, thrive in the Crypto market. RL has been widely applied in various quantitative tasks. However, most methods focus on the spot and could not be directly applied to the futures market with high leverage because of 2 challenges. First, high leverage amplifies reward fluctuations, making training stochastic and difficult to converge. Second, prior works lacked self-awareness of capability boundaries, exposing them to the risk of significant loss when encountering new market state (e.g.,a black swan event like COVID-19). To tackle these challenges, we propose the Efficient and Risk-Aware Ensemble Reinforcement Learning for Futures Trading (FineFT), a novel three-stage ensemble RL framework with stable training and proper risk management. In stage I, ensemble Q learners are selectively updated by ensemble TD errors to improve convergence. In stage II, we filter the Q-learners based on their profitabilities and train VAEs on market states to identify the capability boundaries of the learners. In stage III, we choose from the filtered ensemble and a conservative policy, guided by trained VAEs, to maintain profitability and mitigate risk with new market states. Through extensive experiments on crypto futures in a high-frequency trading environment with high fidelity and 5x leverage, we demonstrate that FineFT outperforms 12 SOTA baselines in 6 financial metrics, reducing risk by more than 40% while achieving superior profitability compared to the runner-up. Visualization of the selective update mechanism shows that different agents specialize in distinct market dynamics, and ablation studies certify routing with VAEs reduces maximum drawdown effectively, and selective update improves convergence and performance.

Shuxin Li, Chang Yang, Youzhi Zhang et al.

Nash equilibrium (NE) is a widely adopted solution concept in game theory due to its stability property. However, we observe that the NE strategy might not always yield the best results, especially against opponents who do not adhere to NE strategies. Based on this observation, we pose a new game-solving question: Can we learn a model that can exploit any, even NE, opponent to maximize their own utility? In this work, we make the first attempt to investigate this problem through in-context learning. Specifically, we introduce a novel method, In-Context Exploiter (ICE), to train a single model that can act as any player in the game and adaptively exploit opponents entirely by in-context learning. Our ICE algorithm involves generating diverse opponent strategies, collecting interactive history training data by a reinforcement learning algorithm, and training a transformer-based agent within a well-designed curriculum learning framework. Finally, comprehensive experimental results validate the effectiveness of our ICE algorithm, showcasing its in-context learning ability to exploit any unknown opponent, thereby positively answering our initial game-solving question.

Rui Ye, Keduan Huang, Qimin Wu et al.

LLM-based multi-agent systems (MAS) have demonstrated significant potential in enhancing single LLMs to address complex and diverse tasks in practical applications. Despite considerable advancements, the field lacks a unified codebase that consolidates existing methods, resulting in redundant re-implementation efforts, unfair comparisons, and high entry barriers for researchers. To address these challenges, we introduce MASLab, a unified, comprehensive, and research-friendly codebase for LLM-based MAS. (1) MASLab integrates over 20 established methods across multiple domains, each rigorously validated by comparing step-by-step outputs with its official implementation. (2) MASLab provides a unified environment with various benchmarks for fair comparisons among methods, ensuring consistent inputs and standardized evaluation protocols. (3) MASLab implements methods within a shared streamlined structure, lowering the barriers for understanding and extension. Building on MASLab, we conduct extensive experiments covering 10+ benchmarks and 8 models, offering researchers a clear and comprehensive view of the current landscape of MAS methods. MASLab will continue to evolve, tracking the latest developments in the field, and invite contributions from the broader open-source community.

Zhiyuan Hu, Shiyun Xiong, Yifan Zhang et al.

Recent advancements in visual language models (VLMs) have notably enhanced their capabilities in handling complex Graphical User Interface (GUI) interaction tasks. Despite these improvements, current frameworks often struggle to generate correct actions in challenging GUI environments. State-of-the-art commercial VLMs are black-boxes, and fine-tuning open-source VLMs for GUI tasks requires significant resources. Additionally, existing trajectory-level evaluation and refinement techniques frequently fall short due to delayed feedback and local optimization issues. To address these challenges, we propose an approach that guides VLM agents with process supervision by a reward model during GUI navigation and control at inference time. This guidance allows the VLM agent to optimize actions at each inference step, thereby improving performance in both static and dynamic environments. In particular, our method demonstrates significant performance gains in three GUI navigation tasks, achieving a 3.4% improvement in single step action accuracy for static environments, along with a around 33% increase in task success rate in one dynamic environment. With further integration of trajectory reflection and retry mechanisms, we also demonstrate even greater enhancement in task success.

Zhen Zhang, Liangcai Su, Zhuo Chen et al.

Deep research agents have achieved remarkable progress on complex information seeking tasks. Even long ReAct style rollouts explore only a single trajectory, while recent state of the art systems scale inference time compute via parallel search and aggregation. Yet deep research answers are composed of complementary pieces of evidence, which parallel rollouts often duplicate rather than complete, yielding diminishing returns while pushing the aggregation context toward the model's limit. We propose Argus, an agentic system in which a Searcher and a Navigator cooperate to treat deep research as assembling a jigsaw from complementary evidence pieces, rather than brute forcing the whole answer in parallel. The Searcher collects evidence traces for a given sub-query through ReAct-style interaction. The Navigator maintains a shared evidence graph, verifying which pieces are still missing, dispatching Searchers to gather them, and reasoning over the completed graph to produce a source-traced final answer. We train the Navigator with reinforcement learning to verify, dispatch, and synthesize, while independently training the Searcher to remain a standard ReAct agent. The resulting Navigator supports rollouts with a single Searcher or many in parallel without retraining. With both Searcher and Navigator built on a 35B-A3B MoE backbone, Argus gains 5.5 points with a single Searcher and 12.7 points with 8 parallel Searchers, averaged over eight benchmarks. With 64 Searchers it reaches 86.2 on BrowseComp, surpassing every proprietary agent we benchmark, while the Navigator's reasoning context stays under 21.5K tokens.

Qingyao Li, Weiwen Liu, Weinan Zhang et al.

Recent advancements in Large Language Models (LLMs) have successfully employed search-based strategies to enhance code generation. However, existing methods typically rely on static, sparse public test cases for verification, leading to pseudo-correctness -- where solutions overfit the visible public tests but fail to generalize to hidden test cases. We argue that optimizing against a fixed, weak environment inherently limits robustness. To address this, we propose AdverMCTS, a novel adversarial Monte Carlo Tree Search framework that combats pseudo-correctness by coupling code search with active vulnerability discovery. AdverMCTS formulates generation as a minimax-style game between a Solver agent, which synthesizes code candidates, and an Attacker agent, which evolves to generate targeted corner test cases that exploit logical divergences in the current code pool. These discovered tests form a dynamic, progressively hostile filter that penalizes fragile reasoning. Extensive experiments demonstrate that AdverMCTS significantly outperforms state-of-the-art baselines, effectively reducing false positive rates and forcing the model to generalize beyond the initial constraints. The resources of this work are available at https://anonymous.4open.science/r/AdverMCTS_open-A255.

Lang Feng, Weihao Tan, Zhiyi Lyu et al.

Online fine-tuning vision-language model (VLM) agents with reinforcement learning (RL) has shown promise for equipping agents with multi-step, goal-oriented capabilities in dynamic environments. However, their open-ended textual action space and non-end-to-end nature of action generation present significant challenges to effective online exploration in RL, e.g., explosion of the exploration space. We propose a novel online fine-tuning method, Counterfactual Soft Reinforcement Learning (CoSo), better suited to the textual output space of VLM agents. Compared to prior methods that assign uniform uncertainty to all tokens, CoSo leverages counterfactual reasoning to dynamically assess the causal influence of individual tokens on post-processed actions. By prioritizing the exploration of action-critical tokens while reducing the impact of semantically redundant or low-impact tokens, CoSo enables a more targeted and efficient online rollout process. We provide theoretical analysis proving CoSo's convergence and policy improvement guarantees, and extensive empirical evaluations supporting CoSo's effectiveness. Our results across a diverse set of agent tasks, including Android device control, card gaming, and embodied AI, highlight its remarkable ability to enhance exploration efficiency and deliver consistent performance gains. The code is available at https://github.com/langfengQ/CoSo.

Lang Feng, Longtao Zheng, Shuo He et al.

Multi-agent LLM systems enable advanced reasoning and tool use via role specialization, yet reliable reinforcement learning (RL) post-training for such systems remains difficult. In this work, we theoretically pinpoint a key reason for training instability when extending group-based RL to multi-agent LLM systems. We show that under GRPO-style optimization, a global normalization baseline may deviate from diverse agents' reward distributions, which ultimately leads to gradient-norm instability. Based on this finding, we propose Dr. MAS, a simple and stable RL training recipe for multi-agent LLM systems. Dr. MAS uses an agent-wise remedy: normalizing advantages per agent using each agent's own reward statistics, which calibrates gradient scales and dramatically stabilizes training, both theoretically and empirically. Beyond the algorithm, Dr. MAS provides an end-to-end RL training framework for multi-agent LLM systems, supporting scalable orchestration, flexible per-agent LLM serving and optimization configs, and shared resource scheduling of LLM actor backends. We evaluate Dr. MAS on multi-agent math reasoning and multi-turn search benchmarks using Qwen2.5 and Qwen3 series models. Dr. MAS achieves clear gains over vanilla GRPO (e.g., +5.6\% avg@16 and +4.6\% pass@16 on math, and +15.2\% avg@16 and +13.1\% pass@16 on search) while largely eliminating gradient spikes. Moreover, it remains highly effective under heterogeneous agent-model assignments while improving efficiency.