Team formation and the dynamics of team-based learning have drawn significant interest in the context of Multi-Agent Reinforcement Learning (MARL). However, existing studies primarily focus on unilateral groupings, predefined teams, or fixed-population settings, leaving the effects of algorithmic bilateral grouping choices in dynamic populations underexplored. To address this gap, we introduce a framework for learning two-sided team formation in dynamic multi-agent systems. Through this study, we gain insight into what algorithmic properties in bilateral team formation influence policy performance and generalization. We validate our approach using widely adopted multi-agent scenarios, demonstrating competitive performance and improved generalization in most scenarios.

The optimization of quality of experience (QoE) in multi-user virtual reality (VR) interactions demands a delicate balance between ultra-low latency, high-fidelity motion synchronization, and equitable resource allocation. While adaptive keyframe extraction mitigates transmission overhead, existing approaches often overlook the causal relationships among allocated bandwidth, CPU frequency, and user perception, limiting QoE gains. This paper proposes an intelligent framework to maximize QoE by integrating adaptive keyframe extraction with causal-aware reinforcement learning (RL). First, a novel QoE metric is formulated using the Weber-Fechner Law, combining perceptual sensitivity, attention-driven priorities, and motion reconstruction accuracy. The QoE optimization problem is then modeled as a mixed integer programming (MIP) task, jointly optimizing keyframe ratios, bandwidth, and computational resources under horizon-fairness constraints. We propose Partial State Causal Deep Deterministic Policy Gradient (PS-CDDPG), which integrates the Deep Deterministic Policy Gradient (DDPG) method with causal influence detection. By leveraging causal information regarding how QoE is influenced and determined by various actions, we explore actions guided by weights calculated from causal inference (CI), which in turn improves training efficiency. Experiments conducted with the CMU Motion Capture Database demonstrate that our framework significantly reduces interactive latency, enhances QoE, and maintains fairness, achieving superior performance compared to benchmark methods.

Bias problems in the estimation of $Q$-values are a well-known obstacle that slows down convergence of $Q$-learning and actor-critic methods. One of the reasons of the success of modern RL algorithms is partially a direct or indirect overestimation reduction mechanism. We propose an easy to implement method built on top of distributional reinforcement learning (DRL) algorithms to deal with the overestimation in a locally adaptive way. Our framework is simple to implement, existing distributional algorithms can be improved with a few lines of code. We provide theoretical evidence and use double $Q$-learning to show how to include locally adaptive overestimation control in existing algorithms. Experiments are provided for tabular, Atari, and MuJoCo environments.

Offline Goal-Conditioned Reinforcement Learning seeks to train agents to reach specified goals from previously collected trajectories. Scaling that promises to long-horizon tasks remains challenging, notably due to compounding value-estimation errors. Principled geometric offers a potential solution to address these issues. Following this insight, we introduce Projective Quasimetric Planning (ProQ), a compositional framework that learns an asymmetric distance and then repurposes it, firstly as a repulsive energy forcing a sparse set of keypoints to uniformly spread over the learned latent space, and secondly as a structured directional cost guiding towards proximal sub-goals. In particular, ProQ couples this geometry with a Lagrangian out-of-distribution detector to ensure the learned keypoints stay within reachable areas. By unifying metric learning, keypoint coverage, and goal-conditioned control, our approach produces meaningful sub-goals and robustly drives long-horizon goal-reaching on diverse a navigation benchmarks.

We introduce Confucius3-Math, an open-source large language model with 14B parameters that (1) runs efficiently on a single consumer-grade GPU; (2) achieves SOTA performances on a range of mathematical reasoning tasks, outperforming many models with significantly larger sizes. In particular, as part of our mission to enhancing education and knowledge dissemination with AI, Confucius3-Math is specifically committed to mathematics learning for Chinese K-12 students and educators. Built via post-training with large-scale reinforcement learning (RL), Confucius3-Math aligns with national curriculum and excels at solving main-stream Chinese K-12 mathematical problems with low cost. In this report we share our development recipe, the challenges we encounter and the techniques we develop to overcome them. In particular, we introduce three technical innovations: Targeted Entropy Regularization, Recent Sample Recovery and Policy-Specific Hardness Weighting. These innovations encompass a new entropy regularization, a novel data scheduling policy, and an improved group-relative advantage estimator. Collectively, they significantly stabilize the RL training, improve data efficiency, and boost performance. Our work demonstrates the feasibility of building strong reasoning models in a particular domain at low cost. We open-source our model and code at https://github.com/netease-youdao/Confucius3-Math.

Particle physics theories, such as those which explain neutrino flavor mixing, arise from a vast landscape of model-building possibilities. A model's construction typically relies on the intuition of theorists. It also requires considerable effort to identify appropriate symmetry groups, assign field representations, and extract predictions for comparison with experimental data. We develop an Autonomous Model Builder (AMBer), a framework in which a reinforcement learning agent interacts with a streamlined physics software pipeline to search these spaces efficiently. AMBer selects symmetry groups, particle content, and group representation assignments to construct viable models while minimizing the number of free parameters introduced. We validate our approach in well-studied regions of theory space and extend the exploration to a novel, previously unexamined symmetry group. While demonstrated in the context of neutrino flavor theories, this approach of reinforcement learning with physics software feedback may be extended to other theoretical model-building problems in the future.

Parallel data collection has redefined Reinforcement Learning (RL), unlocking unprecedented efficiency and powering breakthroughs in large-scale real-world applications. In this paradigm, $N$ identical agents operate in $N$ replicas of an environment simulator, accelerating data collection by a factor of $N$. A critical question arises: \textit{Does specializing the policies of the parallel agents hold the key to surpass the $N$ factor acceleration?} In this paper, we introduce a novel learning framework that maximizes the entropy of collected data in a parallel setting. Our approach carefully balances the entropy of individual agents with inter-agent diversity, effectively minimizing redundancies. The latter idea is implemented with a centralized policy gradient method, which shows promise when evaluated empirically against systems of identical agents, as well as synergy with batch RL techniques that can exploit data diversity. Finally, we provide an original concentration analysis that shows faster rates for specialized parallel sampling distributions, which supports our methodology and may be of independent interest.

Predicting port congestion is crucial for maintaining reliable global supply chains. Accurate forecasts enableimprovedshipment planning, reducedelaysand costs, and optimizeinventoryanddistributionstrategies, thereby ensuring timely deliveries and enhancing supply chain resilience. To achieve accurate predictions, analyzing vessel behavior and their stay times at specific port terminals is essential, focusing particularly on berth scheduling under various conditions. Crucially, the model must capture and learn the underlying priorities and patterns of berth scheduling. Berth scheduling and planning are influenced by a range of factors, including incoming vessel size, waiting times, and the status of vessels within the port terminal. By observing historical Automatic Identification System (AIS) positions of vessels, we reconstruct berth schedules, which are subsequently utilized to determine the reward function via Inverse Reinforcement Learning (IRL). For this purpose, we modeled a specific terminal at the Port of New York/New Jersey and developed Temporal-IRL. This Temporal-IRL model learns berth scheduling to predict vessel sequencing at the terminal and estimate vessel port stay, encompassing both waiting and berthing times, to forecast port congestion. Utilizing data from Maher Terminal spanning January 2015 to September 2023, we trained and tested the model, achieving demonstrably excellent results.

Meta-reinforcement learning requires utilizing prior task distribution information obtained during exploration to rapidly adapt to unknown tasks. The efficiency of an agent's exploration hinges on accurately identifying the current task. Recent Bayes-Adaptive Deep RL approaches often rely on reconstructing the environment's reward signal, which is challenging in sparse reward settings, leading to suboptimal exploitation. Inspired by bisimulation metrics, which robustly extracts behavioral similarity in continuous MDPs, we propose SimBelief-a novel meta-RL framework via measuring similarity of task belief in Bayes-Adaptive MDP (BAMDP). SimBelief effectively extracts common features of similar task distributions, enabling efficient task identification and exploration in sparse reward environments. We introduce latent task belief metric to learn the common structure of similar tasks and incorporate it into the specific task belief. By learning the latent dynamics across task distributions, we connect shared latent task belief features with specific task features, facilitating rapid task identification and adaptation. Our method outperforms state-of-the-art baselines on sparse reward MuJoCo and panda-gym tasks.

Offline reinforcement learning (RL) recently gains growing interests from RL researchers. However, the performance of offline RL suffers from the out-of-distribution problem, which can be corrected by feedback in online RL. Previous offline RL research focuses on restricting the offline algorithm in in-distribution even in-sample action sampling. In contrast, fewer work pays attention to the influence of the batch data. In this paper, we first build a bridge over the batch data and the performance of offline RL algorithms theoretically, from the perspective of model-based offline RL optimization. We draw a conclusion that, with mild assumptions, the distance between the state-action pair distribution generated by the behavioural policy and the distribution generated by the optimal policy, accounts for the performance gap between the policy learned by model-based offline RL and the optimal policy. Secondly, we reveal that in task-agnostic settings, a series of policies trained by unsupervised RL can minimize the worst-case regret in the performance gap. Inspired by the theoretical conclusions, UDG (Unsupervised Data Generation) is devised to generate data and select proper data for offline training under tasks-agnostic settings. Empirical results demonstrate that UDG can outperform supervised data generation on solving unknown tasks.