activity
20242026
collaborators

8 papers

cs.LG2026

The Dark Room in the Reward Channel: Dense Prediction Rewards Collapse GRPO-Trained LLM Agents -- and The Channel, Not the Content, Decides What Works

Yu Wang

Dense per-step supervision is the standard remedy for sparse-reward long-horizon LLM agents: reward the policy for predicting its next observation, which looks provably safe under…

cs.AI2026

Verifiable Process Rewards for Agentic Reasoning

Huining Yuan, Zelai Xu, Huaijie Wang +6

Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning abilities of large language models (LLMs), but most existing approaches rely on sparse outcome-leve…

cs.AI2026

MAGE: Meta-Reinforcement Learning for Language Agents toward Strategic Exploration and Exploitation

Lu Yang, Zelai Xu, Minyang Xie +4

Large Language Model (LLM) agents have demonstrated remarkable proficiency in learned tasks, yet they often struggle to adapt to non-stationary environments with feedback. While In…

cs.CL2026

Translate Policy to Language: Flow Matching Generated Rewards for LLM Explanations

Xinyi Yang, Liang Zeng, Heng Dong +6

As humans increasingly share environments with diverse agents powered by RL, LLMs, and beyond, the ability to explain agent policies in natural language is vital for reliable coexi…

cs.LG2025

Extending Test-Time Scaling: A 3D Perspective with Context, Batch, and Turn

Chao Yu, Qixin Tan, Jiaxuan Gao +7

Reasoning reinforcement learning (RL) has recently revealed a new scaling effect: test-time scaling. Thinking models such as R1 and o1 improve their reasoning accuracy at test time…

cs.AI2025

A Survey on Self-play Methods in Reinforcement Learning

Ruize Zhang, Zelai Xu, Chengdong Ma +8

Self-play, a learning paradigm where agents iteratively refine their policies by interacting with historical or concurrent versions of themselves or other evolving agents, has show…