arXiv Machine Learning

Where the Model Changes Its Mind: Hindsight-Divergence Localization for Efficient Reinforcement Learning with Verifiable Rewards

arXiv Machine Learning
Oct 15

Co-Evolving Acting and Forecasting for Language Agents

The paper introduces CoPE, a method that trains language agents by adding multi‑step action targets to the policy backbone. By supervising each action forecasted from preceding histories, CoPE enables acting and forecasting to co‑evolve without extra inference cost. Across five benchmarks, CoPE outperforms existing RL baselines, achieving up to twice the sample efficiency on ScienceWorld and improving overall performance on realistic applications.

By Haofei Yu, Yixing Chen, Yining Zhao, Xuyan Ye, Yuxiang Liu, Zhigang Hua, Yan Xie, Shuang Yang, Jiaxuan You
arXiv AI
Sep 18

Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL

The paper introduces ActObs, a supervised fine‑tuning method that, unlike standard approaches, also predicts environment observations in agent trajectories. While both ActObs and action‑only training perform similarly after initial fine‑tuning, ActObs diverges during subsequent reinforcement learning, yielding higher pass@k scores on several benchmarks and better cross‑domain task performance. The authors attribute this advantage to ActObs’s joint supervision, which preserves observation gradients and prevents the policy from over‑specializing on actions alone.

By Juzheng Zhang, Disha Makhija, Manoj Ghuhan Arivazhagan, Vinayshekhar Bannihatti Kumar, Rashmi Gangadharaiah
arXiv Machine Learning
Jul 13

Learning More from Less: Reinforcement Learning from Hindsight

arXiv:2607. 09042v1 Announce Type: new Abstract: Reinforcement learning (RL) is increasingly used to post-train vision-language-action (VLA) models, but every update consumes robot rollouts that are slow and costly to collect, making sample efficiency a central concern.

By Iris Xu, Sunshine Jiang, John Marangola, Nitish Dashora, Richard Li, Thomas Liu, Zexue He, Yuheng Zhi, Alex Pentland, Pulkit Agrawal, Zhang-Wei Hong
arXiv AI
Jul 20

Process Reward Informed Tree Rollout for Effective Multi-Turn RL

arXiv:2607. 15610v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become a key approach for training LLM agents, yet popular methods such as GRPO/RLOO rely on multiple independently sampled complete trajectories for advantage estimation.

By Xintong Li, Sha Li, Yuwei Zhang, Changlong Yu, Rongmei Lin, Hongye Jin, Shuyi Guan, Xin Liu, Linwei Li, Qingyu Yin, Jingbo Shang
arXiv AI
Nov 6

AttnSR: Attention Step Reward for Long-Horizon Agentic RL

AttnSR introduces a step-level reward augmentation technique for long-horizon language model agents, deriving reward allocations from policy attention during training replay without extra reward models or rollouts. By tracing backward through the policy attention graph, it estimates the relevance of earlier steps to the final outcome, using the terminal step relative to a baseline to reinforce or penalize selected steps. Experiments on Qwen2.5-7B-Instruct show AttnSR consistently outperforms PPO on ALFWorld, raising held‑out success from 91.2% to 97.1%, and it also improves ScienceWorld while reducing environment interactions by 29.9%.

By Mengsong Wu, Tianchen Zhou, Zhaochen Su, Zhenhua Liu, Zhe Cui, Aimin Zhou
arXiv Machine Learning
Oct 12

TreeHCA: Hindsight-Guided Exploration and Credit Assignment for Multi-Turn LLM Agents

TreeHCA introduces a method that couples hindsight-guided exploration with hindsight credit assignment for multi-turn large language model agents. By using changes in reference-answer likelihood to revisit intermediate histories and sample alternative continuations, it builds a tree of reasoning and tool interactions, and applies an inverse-hindsight importance correction to propagate terminal outcomes into turn-level advantages. Experiments on seven question-answering benchmarks with three Qwen backbones show that TreeHCA outperforms all baselines in aggregate exact match.

By Debajoy Mukherjee, Jeremy Carleton, Sushil Vemuri, Srinivas Shakkottai, Dileep Kalathil