NeurIPS 2025 Papers

Real-world reinforcement learning demands adaptation to unseen environmental conditions without costly retraining. Contextual Markov Decision Processes (cMDP) model this challenge, but existing methods often require explicit context variables (e.g., friction, gravity), limiting their use when contexts are latent or hard to measure. We introduce Dynamics-Aligned Latent Imagination (DALI), a framework integrated within the Dreamer architecture that infers latent context representations from agent-environment interactions. By training a self-supervised encoder to predict forward dynamics, DALI generates actionable representations conditioning the world model and policy, bridging perception and control. We theoretically prove this encoder is essential for efficient context inference and robust generalization. DALI’s latent space enables counterfactual consistency: Perturbing a gravity-encoding dimension alters imagined rollouts in physically plausible ways. On challenging cMDP benchmarks, DALI achieves significant gains over context-unaware baselines, often surpassing context-aware baselines in extrapolation tasks, enabling zero-shot generalization to unseen contextual variations.

Meta-Black-Box Optimization (MetaBBO) streamlines the automation of optimization algorithm design through meta-learning. It typically employs a bi-level structure: the meta-level policy undergoes meta-training to reduce the manual effort required in developing algorithms for low-level optimization tasks. The original MetaBox (2023) provided the first open-source framework for reinforcement learning-based single-objective MetaBBO. However, its relatively narrow scope no longer keep pace with the swift advancement in this field. In this paper, we introduce MetaBox-v2 (\url{https://github.com/MetaEvo/MetaBox}) as a milestone upgrade with four novel features: 1) a unified architecture supporting RL, evolutionary, and gradient-based approaches, by which we reproduce $23$ up-to-date baselines; 2) efficient parallelization schemes, which reduce the training/testing time by $10-40$x; 3) a comprehensive benchmark suite of $18$ synthetic/realistic tasks ($1900$+ instances) spanning single-objective, multi-objective, multi-model, and multi-task optimization scenarios; 4) plentiful and extensible interfaces for custom analysis/visualization and integrating to external optimization tools/benchmarks. To show the utility of MetaBox-v2, we carry out a systematic case study that evaluates the built-in baselines in terms of the optimization performance, generalization ability and learning efficiency. Valuable insights are concluded from thorough and detailed analysis for practitioners and those new to the field.

Inverse Reinforcement Learning (IRL) learns a reward function to explain expert demonstrations. Modern IRL methods often use the adversarial (minimax) formulation that alternates between reward and policy optimization, which often lead to {\em unstable} training. Recent non-adversarial IRL approaches improve stability by jointly learning reward and policy via energy-based formulations but lack formal guarantees. This work bridges this gap. We first present a *unified* view showing canonical non-adversarial methods explicitly or implicitly maximize the likelihood of expert behavior, which is equivalent to minimizing the expected return gap. This insight leads to our main contribution: *Trust Region Reward Optimization* (TRRO), a framework that guarantees *monotonic* improvement in this likelihood via a Minorization-Maximization process. We instantiate TRRO into *Proximal Inverse Reward Optimization* (PIRO), a practical and stable IRL algorithm. Theoretically, TRRO provides the IRL counterpart to the stability guarantees of Trust Region Policy Optimization (TRPO) in forward RL. Empirically, PIRO matches or surpasses state-of-the-art baselines in reward recovery, policy imitation with high sample efficiency on MuJoCo and Gym-Robotics benchmarks and a real-world animal behavior modeling task.

In networked multi-agent reinforcement learning (Networked-MARL), decentralized agents must act autonomously under local observability and constrained communication over fixed physical graphs. Existing methods often assume static neighborhoods, limiting adaptability to dynamic or heterogeneous environments. While centralized frameworks can learn dynamic graphs, their reliance on global state access and centralized infrastructure is impractical in real-world decentralized systems. We propose a stochastic graph-based policy for Networked-MARL, where each agent conditions its decision on a sampled subgraph over its local physical neighborhood. Building on this formulation, we introduce \textbf{BayesG}, a decentralized actor–critic framework that learns sparse, context-aware interaction structures via Bayesian variational inference. Each agent operates over an ego-graph and samples a latent communication mask to guide message passing and policy computation. The variational distribution is trained end-to-end alongside the policy using an evidence lower bound (ELBO) objective, enabling agents to jointly learn both interaction topology and decision-making strategies. BayesG outperforms strong MARL baselines on large-scale traffic control tasks with up to 167 agents, demonstrating superior scalability, efficiency, and performance.

A longstanding goal in safe reinforcement learning (RL) is a method to ensure the safety of a policy throughout the entire process, from learning to operation. However, existing safe RL paradigms inherently struggle to achieve this objective. We propose a method, called Provably Lifetime Safe RL (PLS), that integrates offline safe RL with safe policy deployment to address this challenge. Our proposed method learns a policy offline using return-conditioned supervised learning and then deploys the resulting policy while cautiously optimizing a limited set of parameters, known as target returns, using Gaussian processes (GPs). Theoretically, we justify the use of GPs by analyzing the mathematical relationship between target and actual returns. We then prove that PLS finds near-optimal target returns while guaranteeing safety with high probability. Empirically, we demonstrate that PLS outperforms baselines both in safety and reward performance, thereby achieving the longstanding goal to obtain high rewards while ensuring the safety of a policy throughout the lifetime from learning to operation.

AlphaZero has achieved remarkable success in complex decision-making problems through self-play and neural network training. However, its self-play process remains inefficient due to limited exploration of high-uncertainty positions, the overlooked runner-up decisions in Monte Carlo Tree Search (MCTS), and high variance in value labels. To address these challenges, we propose and evaluate uncertainty-guided exploration by branching from high-uncertainty positions using our proposed Label Change Rate (LCR) metric, which is further refined by a Bayesian inference framework. Our proposed approach leverages runner-up MCTS decisions to create multiple variations, and ensembles value labels across these variations to reduce variance. We investigate three key design parameters for our branching strategy: where to branch, how many variations to branch, and which move to play in the new branch. Our empirical findings indicate that branching with 10 variations per game provides the best performance-exploration balance. Overall, our end-to-end results show an improved sample efficiency over the baseline by 58.5\% on 9x9 Go in the early stage of training and by 47.3\% on 19x19 Go in the late stage of training.

Reinforcement fine-tuning (RFT) has shown great promise in achieving humanlevel reasoning capabilities of Large Language Models (LLMs), and has recently been extended to MLLMs. Nevertheless, reasoning about videos, which is a fundamental aspect of human intelligence, remains a persistent challenge due to the complex logic, temporal and causal structures inherent in video data. To fill this gap, we propose VideoRFT, a novel approach that extends the RFT paradigm to cultivate human-like video reasoning capabilities in MLLMs. VideoRFT follows the standard two-stage scheme in RFT: supervised fine-tuning (SFT) with chain-of-thought (CoT) annotations, followed by reinforcement learning (RL) to improve generalization. A central challenge to achieve this in the video domain lies in the scarcity of large-scale, high-quality video CoT datasets. We address this by building a multi-expert-driven, cognition-inspired CoT curation pipeline. First, we devise a cognition-inspired prompting strategy to elicit a reasoning LLM to generate preliminary CoTs based solely on rich, structured, and literal representations of video content. Subsequently, these CoTs are revised by a MLLM conditioned on the actual video, ensuring visual consistency and reducing visual hallucinations. This pipeline results in two new datasets, i.e.VideoRFT-CoT-102K for SFT and VideoRFT-RL-310K for RL. To further strengthen the RL phase, we introduce a novel semantic-consistency reward that explicitly promotes the alignment between textual reasoning and visual evidence. This reward encourages the model to produce coherent, context-aware reasoning outputs grounded in visual input. Extensive experiments show that VideoRFT achieves state-of-the-art performance on six video reasoning benchmarks.

Code generation with large language models (LLMs), often termed vibe coding, is increasingly adopted in production but fails to ensure code quality, particularly in security (e.g., SQL injection vulnerabilities) and maintainability (e.g., missing type annotations). Existing methods, such as supervised fine-tuning and rule-based post-processing, rely on labor-intensive annotations or brittle heuristics, limiting their scalability and effectiveness. We propose REAL (Reinforcement rEwards from Automated anaLysis), a reinforcement learning framework that trains LLMs to generate production-quality code using program analysis–guided feedback. Specifically, REAL integrates two automated signals: (1) static analyzers detecting security and maintainability defects and (2) unit tests ensuring functional correctness. Unlike prior work, our framework is prompt-agnostic and reference-free, enabling scalable supervision without manual intervention. Experiments across multiple datasets and model scales demonstrate that REAL outperforms state-of-the-art methods in simultaneous assessments of functionality and code quality. Our work bridges the gap between rapid prototyping and production-ready code, enabling LLMs to deliver both speed and quality.

The success of computer vision tasks is mainly attributed to the architectural design of neural networks. This highlights the need to automatically design high-performance architectures via Neural Architecture Search (NAS). To accelerate the search process, training-free NAS is proposed, which aims to search high-performance architectures at initialization via zero-cost proxies (ZCPs). However, existing zero-cost proxies heavily rely on manual design, which is often labor-intensive and requires extensive expert knowledge. In addition, these crafted proxies often suffer from poor correlation with final model performance and high computational complexity, severely limiting NAS efficiency in real-world applications. To address those issues, this paper proposes a novel Large Language Models (LLMs)-driven $\underline{A}$utomatic $\underline{P}$roxy $\underline{D}$iscovery ($\textbf{APD}$) framework, which revolutionizes the design paradigm of ZCPs by leveraging LLMs to automatically discover optimal ZCPs for Training-Free NAS. Moreover, we utilize actor-critic based reinforcement learning to optimize prompts, enabling to generate better ZCPs in the next generation. We conduct extensive experiments on mainstream NAS benchmarks, demonstrating APD excels in both performance and efficiency. Besides, we firmly believe that our APD will dramatically benefit the deep learning community through providing novel paradigm of design algorithms via LLMs.

Linear TD($\lambda$) is one of the most fundamental reinforcement learning algorithms for policy evaluation. Previously, convergence rates are typically established under the assumption of linearly independent features, which does not hold in many practical scenarios. This paper instead establishes the first $L^2$ convergence rates for linear TD($\lambda$) operating under arbitrary features, without making any algorithmic modification or additional assumptions. Our results apply to both the discounted and average-reward settings. To address the potential non-uniqueness of solutions resulting from arbitrary features, we develop a novel stochastic approximation result featuring convergence rates to the solution set instead of a single point.

Pre-training vision-language representations on human action videos has emerged as a promising approach to reduce reliance on large-scale expert demonstrations for training embodied agents. However, prior methods often employ time con- trastive learning based on goal-reaching heuristics, progressively aligning language instructions from the initial to the final frame. This overemphasis on future frames can result in erroneous vision-language associations, as actions may terminate early or include irrelevant moments in the end. To address this issue, we propose Action Temporal Coherence Learning (AcTOL) to learn ordered and continuous vision-language representations without rigid goal-based constraint. AcTOL treats a video as a continuous trajectory where it (1) contrasts semantic differences be- tween frames to reflect their natural ordering, and (2) imposes a local Brownian bridge constraint to ensure smooth transitions across intermediate frames. Exten- sive imitation learning experiments on both simulated and real robots show that the pretrained features significantly enhance downstream manipulation tasks with high robustness to different linguistic styles of instructions, offering a viable pathway toward generalized embodied agents. Our project page is at https://actol-pretrain.github.io/.

Current transformer-based imitation learning approaches introduce discrete action representations and train an autoregressive transformer decoder on the resulting latent code. However, the initial quantization breaks the continuous structure of the action space thereby limiting the capabilities of the generative model. We propose a quantization-free method instead that leverages Generative Infinite-Vocabulary Transformers (GIVT) as a direct, continuous policy parametrization for autoregressive transformers. This simplifies the imitation learning pipeline while achieving state-of-the-art performance on a variety of popular simulated robotics tasks. We enhance our policy roll-outs by carefully studying sampling algorithms, further improving the results.

Effective robotic exploration in continuous domains requires planning trajectories that maximize coverage over a predefined region. A recent development, Stein Variational Ergodic Search (SVES), proposed parallel ergodic exploration (a key approach within the field of robotic exploration), via Stein variational inference that computes a set of candidate trajectories approximating the posterior distribution over the solution space trajectories. While this approach leverages GPU parallelism well, the trajectories in the set might not be distinct enough, leading to a suboptimal set. In this paper, we propose two key methods to diversify the solution set of this approach. First, we leverage the signature kernel within the SVES framework, introducing a pathwise, sequence-sensitive interaction that preserves the Markovian structure of the trajectories and naturally spreads paths across distinct regions of the search space. Second, we propose a derivative-free evolution-strategy interpretation of SVES that exploits batched, GPU-friendly fitness evaluations and can be paired with approximate gradients whenever analytic gradients of the kernel are unavailable or computationally intractable. The resulting method both retains SVES’s advantages while diversifying the solution set and extending its reach to black-box objectives. Across planar forest search, 3D quadrotor coverage, and model-predictive control benchmarks, our approach consistently reduces ergodic cost and produces markedly richer trajectory sets than SVES without significant extra tuning effort.

Process reward model (PRM) has been proven effective in test-time scaling of LLM on challenging reasoning tasks. However, the reward hacking induced by PRM hinders its successful applications in reinforcement fine-tuning. We find the primary cause of reward hacking induced by PRM is that: the canonical summation-form credit assignment in reinforcement learning (RL), i.e. cumulative gamma-decayed future rewards, causes the LLM to hack steps with high rewards. Therefore, to unleashing the power of PRM in training-time, we propose PURE: Process sUpervised Reinforcement lEarning. The core of PURE is the min-form credit assignment that defines the value function as the minimum future rewards. This method unifies the optimization objective with respect to process rewards during test-time and training-time, and significantly alleviates reward hacking due to the limits on the range of values of value function and more rational assignment of advantages. Through extensively experiments on 3 base models, we achieve similar reasoning performance using PRM-based approach compared with verifiable reward-based approach if enabling min-form credit assignment. In contrast, the canonical sum-form credit assignment even collapses training at the beginning. Moreover, when we incorporate 1/10th verifiable rewards to auxiliary the PRM-based fine-tuning, it further alleviate reward hacking and results in the best fine-tuned model based on Qwen2.5-Math-7B with 82.5% accuracy on AMC23 and 53.3% average accuracy across 5 benchmarks. Furthermore, we summary the reward hacking cases we encountered during training and analysis the cause of training collapse.

Constrained decision-making is essential for designing safe policies in real-world control systems, yet simulated environments often fail to capture real-world adversities. We consider the problem of learning a policy that will maximize the cumulative reward while satisfying a constraint, even when there is a mismatch between the real model and an accessible simulator/nominal model. In particular, we consider the robust constrained Markov decision problem (RCMDP) where an agent needs to maximize the reward and satisfy the constraint against the worst possible stochastic model under the uncertainty set centered around an unknown nominal model. Primal-dual methods, effective for standard constrained MDP (CMDP), are not applicable here because of the lack of the strong duality property. Further, one cannot apply the standard robust value-iteration based approach on the composite value function, either, as the worst-case models may be different for the reward value function and the constraint value function. We propose a novel technique that effectively minimizes the constraint value function--to satisfy the constraints; on the other hand, when all the constraints are satisfied, it can simply maximize the robust reward value function. We prove that such an algorithm finds a policy with at most $\epsilon$ sub-optimality and a feasible policy after $O(\epsilon^{-2})$ iterations. In contrast to the state-of-the-art method, we do not need to employ a binary search; thus, we reduce the computation time and achieve a better performance, especially for continuous state-space.

Binary choices, as often used for reinforcement learning from human feedback (RLHF), convey only the *direction* of a preference. A person may choose apples over oranges and bananas over grapes, but *which preference is stronger*? Strength is crucial for decision-making under uncertainty and generalization of preference models, but hard to measure reliably. Metadata such as response times and inter-annotator agreement can serve as proxies for strength, but are often noisy and confounded. We propose ResponseRank to address the challenge of learning from noisy strength signals. Our method uses relative differences in these signals to *rank responses to pairwise comparisons by their inferred preference strength*. Signals are only considered locally within carefully constructed strata, controlling for systemic variation. This enables robust learning of utility differences consistent with strength-derived rankings, all while making minimal assumptions. Our contributions are threefold: (1) ResponseRank, a novel method that robustly learns preference strength by leveraging locally valid relative strength signals; (2) empirical evidence of improved sample efficiency and robustness across diverse tasks: synthetic preference learning (with simulated response times), language modeling (with annotator agreement), and RL control tasks (with simulated episode returns); and (3) the *Pearson Distance Correlation (PDC)*, a novel metric that isolates cardinal utility learning from ordinal accuracy.

Vision-language models have been explored for radiology report generation with promising results. Yet, uncertainty elaborated in findings and the reasoning process for reaching clinical impressions are seldom explicitly modeled, reducing the clinical accuracy and trustworthiness of the generated reports. We present CURV, a novel framework that alleviates the limitations through integrated awareness of uncertainty and explicit reasoning capabilities. Our approach consists of three key components: (1) an uncertainty modeling mechanism that teaches the model to recognize and express appropriate levels of diagnostic confidence, (2) a structured reasoning framework that generates intermediate explanatory steps connecting visual findings to clinical impressions, and (3) a reasoning coherence reward that ensures logical consistency among findings, reasoning, and impressions. We implement CURV through a three-stage training pipeline that combines uncertainty-aware fine-tuning, reasoning initialization, and reinforcement learning. In particular, we adopt a comprehensive reward function addresses multiple aspects of report quality, incorporating medical term matching, uncertainty expression evaluation, and semantic coherence evaluation. Experimental results demonstrate that CURV generates clinically relevant reports with appropriate uncertainty expressions and transparent reasoning traces, significantly outperforming previous methods. CURV represents a substantial advancement toward interpretable and trustworthy AI-generated radiology reports, with broader implications for the deployment of vision-language models in high-stakes clinical environments where uncertainty awareness and reasoning transparency are essential.

Retrosynthetic planning is a fundamental task in chemical discovery. Due to the vast combinatorial search space, identifying viable synthetic routes remains a significant challenge--even for expert chemists. Recent advances in Large Language Models (LLMs), particularly equipped with reinforcement learning, have demonstrated strong human-like reasoning and planning abilities, especially in mathematics and code problem solving. This raises a natural question: Can the reasoning capabilities of LLMs be harnessed to develop an AI chemist capable of learning effective policies for multi-step retrosynthesis? In this study, we introduce Retro-R1, a novel LLM-based retrosynthesis agent trained via reinforcement learning to design molecular synthesis pathways. Unlike prior approaches, which typically rely on single-turn, question-answering formats, Retro-R1 interacts dynamically with plug-in single-step retrosynthesis tools and learns from environmental feedback. Experimental results show that Retro-R1 achieves a 55.79\% pass@1 success rate, surpassing the previous state of the art by 8.95\%. Notably, Retro-R1 demonstrates strong generalization to out-of-domain test cases, where existing methods tend to fail despite their high in-domain performance. Our work marks a significant step toward equipping LLMs with advanced, chemist-like reasoning abilities, highlighting the promise of reinforcement learning for enabling data-efficient, generalizable, and sophisticated scientific problem-solving in LLM-based agents.

Options markets represent one of the most sophisticated segments of the financial ecosystem, with prices that directly reflect market uncertainty. In this paper, we introduce the first reinforcement learning (RL) framework specifically designed for volatility trading through options, focusing on profit from the difference between implied volatility and realized volatility. Our multi-agent architecture consists of an Option Position Agent (OP-Agent) responsible for volatility timing by controlling long/short volatility positions, and a Hedger Routing Agent (HR-Agent) that manages risk and maximizes path-dependent profits by selecting optimal hedging strategies with different risk preferences. Evaluating our approach using cryptocurrency options data from 2021-2024, we demonstrate superior performance on BTC and ETH, significantly outperforming traditional strategies and machine learning baselines across all profit and risk-adjusted metrics while exhibiting sophisticated trading behavior. The code framework and sample data of this paper have been released on https://github.com/Edwicn/OPHR-MasteringVolatilityTradingwithMultiAgentDeepReinforcementLearning