Yifan Wu

Zhaorun Chen, Zhuokai Zhao, Kai Zhang et al.

While reinforcement learning (RL) can empower autonomous agents by enabling self-improvement through interaction, its practical adoption remains challenging due to costly rollouts, limited task diversity, unreliable reward signals, and infrastructure complexity, all of which obstruct the collection of scalable experience data. To address these challenges, we introduce DreamGym, the first unified framework designed to synthesize diverse experiences with scalability in mind to enable effective online RL training for autonomous agents. Rather than relying on expensive real-environment rollouts, DreamGym distills environment dynamics into a reasoning-based experience model that derives consistent state transitions and feedback signals through step-by-step reasoning, enabling scalable agent rollout collection for RL. To improve the stability and quality of transitions, DreamGym leverages an experience replay buffer initialized with offline real-world data and continuously enriched with fresh interactions to actively support agent training. To improve knowledge acquisition, DreamGym adaptively generates new tasks that challenge the current agent policy, enabling more effective online curriculum learning. Experiments across diverse environments and agent backbones demonstrate that DreamGym substantially improves RL training, both in fully synthetic settings and in sim-to-real transfer scenarios. On non-RL-ready tasks like WebArena, DreamGym outperforms all baselines by over 30%. And in RL-ready but costly settings, it matches GRPO and PPO performance using only synthetic interactions. When transferring a policy trained purely on synthetic experiences to real-environment RL, DreamGym yields significant additional performance gains while requiring far fewer real-world interactions, providing a scalable warm-start strategy for general-purpose RL.

Kai Zhang, Xiangchao Chen, Bo Liu et al. · microsoft-research

A long-term goal of language agents is to learn and improve through their own experience, ultimately outperforming humans in complex, real-world tasks. However, training agents from experience data with reinforcement learning remains difficult in many environments, which either lack verifiable rewards (e.g., websites) or require inefficient long-horizon rollouts (e.g., multi-turn tool use). As a result, most current agents rely on supervised fine-tuning on expert data, which is challenging to scale and generalizes poorly. This limitation stems from the nature of expert demonstrations: they capture only a narrow range of scenarios and expose the agent to limited environment diversity. We address this limitation with a middle-ground paradigm we call early experience: interaction data generated by the agent's own actions, where the resulting future states serve as supervision without reward signals. Within this paradigm we study two strategies of using such data: (1) Implicit world modeling, which uses collected states to ground the policy in environment dynamics; and (2) Self-reflection, where the agent learns from its suboptimal actions to improve reasoning and decision-making. We evaluate across eight diverse environments and multiple model families. Our approaches consistently improve effectiveness and out-of-domain generalization, highlighting the value of early experience. Moreover, in environments with verifiable rewards, our results provide promising signals that early experience offers a strong foundation for subsequent reinforcement learning, positioning it as a practical bridge between imitation learning and fully experience-driven agents.

Yifan Wu, Xichen Ye, Songmin Dai et al.

Out-of-distribution (OOD) detection is an essential approach to robustifying deep learning models, enabling them to identify inputs that fall outside of their trained distribution. Existing OOD detection methods usually depend on crafted data, such as specific outlier datasets or elaborate data augmentations. While this is reasonable, the frequent mismatch between crafted data and OOD data limits model robustness and generalizability. In response to this issue, we introduce Outlier Exposure by Simple Transformations (OEST), a framework that enhances OOD detection by leveraging "peripheral-distribution" (PD) data. Specifically, PD data are samples generated through simple data transformations, thus providing an efficient alternative to manually curated outliers. We adopt energy-based models (EBMs) to study PD data. We recognize the "energy barrier" in OOD detection, which characterizes the energy difference between in-distribution (ID) and OOD samples and eases detection. PD data are introduced to establish the energy barrier during training. Furthermore, this energy barrier concept motivates a theoretically grounded energy-barrier loss to replace the classical energy-bounded loss, leading to an improved paradigm, OEST*, which achieves a more effective and theoretically sound separation between ID and OOD samples. We perform empirical validation of our proposal, and extensive experiments across various benchmarks demonstrate that OEST* achieves better or similar accuracy compared with state-of-the-art methods.