Large Language Models (LLMs) exhibit remarkable code generation capabilities but falter when adapting to frequent updates in external library APIs. This critical limitation, stemming from reliance on outdated API knowledge from their training data, even with access to current documentation, impedes reliable code generation in dynamic environments. To tackle this issue, we propose ReCode (rule-based Reinforcement learning for Code Update), a novel framework that mimics human programmer adaptation to API changes. Specifically, we construct a dataset of approximately 2,000 data entries to train the LLMs to perform version migration based on updated information. Then, we introduce a modified string similarity metric for code evaluation as the reward for reinforcement learning. Our experiments demonstrate that ReCode substantially boosts LLMs' code generation performance in dynamic API scenarios, especially on the unseen CodeUpdateArena task. Crucially, compared to supervised fine-tuning, ReCode has less impact on LLMs' general code generation abilities. We apply ReCode on various LLMs and reinforcement learning algorithms (GRPO and DAPO), all achieving consistent improvements. Notably, after training, Qwen2.5-Coder-7B outperforms that of the 32B parameter code instruction-tuned model and the reasoning model with the same architecture. Code is available at https://github.com/zjunlp/ReCode.

This paper presents a real time, data driven decision support framework for epidemic control. We combine a compartmental epidemic model with sequential Bayesian inference and reinforcement learning (RL) controllers that adaptively choose intervention levels to balance disease burden, such as intensive care unit (ICU) load, against socio economic costs. We construct a context specific cost function using empirical experiments and expert feedback. We study two RL policies: an ICU threshold rule computed via Monte Carlo grid search, and a policy based on a posterior averaged Q learning agent. We validate the framework by fitting the epidemic model to publicly available ICU occupancy data from the COVID 19 pandemic in England and then generating counterfactual roll out scenarios under each RL controller, which allows us to compare the RL policies to the historical government strategy. Over a 300 day period and for a range of cost parameters, both controllers substantially reduce ICU burden relative to the observed interventions, illustrating how Bayesian sequential learning combined with RL can support the design of epidemic control policies.

Multi-agent reinforcement learning (MARL) has been increasingly adopted in many real-world applications. While MARL enables decentralized deployment on resource-constrained edge devices, it suffers from severe non-stationarity due to the synchronous updates of agent policies. This non stationarity results in unstable training and poor policy con vergence, especially as the number of agents increases. In this paper, we propose RELED, a scalable MARL framework that integrates large language model (LLM)-driven expert demonstrations with autonomous agent exploration. RELED incorporates a Stationarity-Aware Expert Demonstration module, which leverages theoretical non-stationarity bounds to enhance the quality of LLM-generated expert trajectories, thus providing high reward and training-stable samples for each agent. Moreover, a Hybrid Expert-Agent Policy Optimization module adaptively balances each agent's learning from both expert-generated and agent-generated trajectories, accelerating policy convergence and improving generalization. Extensive experiments with real city networks based on OpenStreetMap demonstrate that RELED achieves superior performance compared to state-of-the-art MARL methods.

Advertising (Ad) is a cornerstone of the digital economy, yet the moderation of video advertisements remains a significant challenge due to their complexity and the need for precise violation localization. While recent advancements, such as the RAVEN model, have improved coarse-grained violation detection, critical gaps persist in fine-grained understanding, explainability, and generalization. To address these limitations, we propose RAVEN++, a novel framework that introduces three key innovations: 1) Active Reinforcement Learning (RL), which dynamically adapts training to samples of varying difficulty; 2) Fine-Grained Violation Understanding, achieved through hierarchical reward functions and reasoning distillation; and 3) Progressive Multi-Stage Training, which systematically combines knowledge injection, curriculum-based passive RL, and active RL. Extensive experiments on both public and proprietary datasets, on both offline scenarios and online deployed A/B Testing, demonstrate that RAVEN++ outperforms general-purpose LLMs and specialized models like RAVEN in terms of fine-grained violation understanding, reasoning capabilities, and generalization ability.

We propose a refinement of temporal-difference learning that enforces first-order Bellman consistency: the learned value function is trained to match not only the Bellman targets in value but also their derivatives with respect to states and actions. By differentiating the Bellman backup through differentiable dynamics, we obtain analytically consistent gradient targets. Incorporating these into the critic objective using a Sobolev-type loss encourages the critic to align with both the value and local geometry of the target function. This first-order TD matching principle can be seamlessly integrated into existing algorithms, such as Q-learning or actor-critic methods (e.g., DDPG, SAC), potentially leading to faster critic convergence and more stable policy gradients without altering their overall structure.

Existing reinforcement learning methods for Chain-of-Thought reasoning suffer from two critical limitations. First, they operate as monolithic black boxes that provide undifferentiated reward signals, obscuring individual step contributions and hindering error diagnosis. Second, sequential decoding has O(n) time complexity. This makes real-time deployment impractical for complex reasoning tasks. We present DeCoRL (Decoupled Reasoning Chains via Coordinated Reinforcement Learning), a novel framework that transforms reasoning from sequential processing into collaborative modular orchestration. DeCoRL trains lightweight specialized models to generate reasoning sub-steps concurrently, eliminating sequential bottlenecks through parallel processing. To enable precise error attribution, the framework designs modular reward functions that score each sub-step independently. Cascaded DRPO optimization then coordinates these rewards while preserving inter-step dependencies. Comprehensive evaluation demonstrates state-of-the-art results across RM-Bench, RMB, and RewardBench, outperforming existing methods including large-scale models. DeCoRL delivers 3.8 times faster inference while maintaining superior solution quality and offers a 22.7\% improvement in interpretability through explicit reward attribution. These advancements, combined with a 72.4\% reduction in energy consumption and a 68\% increase in throughput, make real-time deployment of complex reasoning systems a reality.

The challenge of building neural networks that can continuously learn and adapt to evolving data streams is central to the fields of continual learning (CL) and reinforcement learning (RL). This lifelong learning problem is often framed in terms of the plasticity-stability dilemma, focusing on issues like loss of plasticity and catastrophic forgetting. Unlike neural networks, biological brains maintain plasticity through capacity growth, inspiring researchers to explore similar approaches in artificial networks, such as adding capacity dynamically. Prior solutions often lack parameter efficiency or depend on explicit task indices, but Mixture-of-Experts (MoE) architectures offer a promising alternative by specializing experts for distinct distributions. This paper aims to evaluate a DynamicMoE approach for continual and reinforcement learning environments and benchmark its effectiveness against existing network expansion methods.

Group-based policy optimization methods like GRPO and GSPO have become standard for training multimodal models, leveraging group-wise rollouts and relative advantage estimation. However, they suffer from a critical \emph{gradient vanishing} problem when all responses within a group receive identical rewards, causing advantage estimates to collapse and training signals to diminish. Existing attempts to mitigate this issue fall into two paradigms: filtering-based and sampling-based methods. Filtering-based methods first generate rollouts broadly and then retroactively filter out uninformative groups, leading to substantial computational overhead. Sampling-based methods proactively select effective samples before rollout but rely on static criteria or prior dataset knowledge, lacking real-time adaptability. To address these issues, we propose \textbf{VADE}, a \textbf{V}ariance-\textbf{A}ware \textbf{D}ynamic sampling framework via online sample-level difficulty \textbf{E}stimation. Our framework integrates three key components: online sample-level difficulty estimation using Beta distributions, a Thompson sampler that maximizes information gain through the estimated correctness probability, and a two-scale prior decay mechanism that maintains robust estimation under policy evolution. This three components design enables VADE to dynamically select the most informative samples, thereby amplifying training signals while eliminating extra rollout costs. Extensive experiments on multimodal reasoning benchmarks show that VADE consistently outperforms strong baselines in both performance and sample efficiency, while achieving a dramatic reduction in computational overhead. More importantly, our framework can serves as a plug-and-play component to be seamlessly integrated into existing group-based RL algorithms. Code and models are available at https://VADE-RL.github.io.

In this work, we investigate how implicit neural feed back can accelerate reinforcement learning in complex robotic manipulation settings. While prior electroencephalogram (EEG) guided reinforcement learning studies have primarily focused on navigation or low-dimensional locomotion tasks, we aim to understand whether such neural evaluative signals can improve policy learning in high-dimensional manipulation tasks involving obstacles and precise end-effector control. We integrate error related potentials decoded from offline-trained EEG classifiers into reward shaping and systematically evaluate the impact of human-feedback weighting. Experiments on a 7-DoF manipulator in an obstacle-rich reaching environment show that neural feedback accelerates reinforcement learning and, depending on the human-feedback weighting, can yield task success rates that at times exceed those of sparse-reward baselines. Moreover, when applying the best-performing feedback weighting across all sub jects, we observe consistent acceleration of reinforcement learning relative to the sparse-reward setting. Furthermore, leave-one subject-out evaluations confirm that the proposed framework remains robust despite the intrinsic inter-individual variability in EEG decodability. Our findings demonstrate that EEG-based reinforcement learning can scale beyond locomotion tasks and provide a viable pathway for human-aligned manipulation skill acquisition.

The transition to autonomous material systems necessitates adaptive control methodologies to maximize structural longevity. This study frames the self-healing process as a Reinforcement Learning (RL) problem within a Markov Decision Process (MDP), enabling agents to autonomously derive optimal policies that efficiently balance structural integrity maintenance against finite resource consumption. A comparative evaluation of discrete-action (Q-learning, DQN) and continuous-action (TD3) agents in a stochastic simulation environment revealed that RL controllers significantly outperform heuristic baselines, achieving near-complete material recovery. Crucially, the TD3 agent utilizing continuous dosage control demonstrated superior convergence speed and stability, underscoring the necessity of fine-grained, proportional actuation in dynamic self-healing applications.