paper-with-me

Papers

When Denser Credit Is Not Enough: Evidence-Calibrated Policy Optimization for Long-Horizon LLM Agent Training

2026-06-04 · Yuanfan Li, Qi Zhou, Wenjing Duan, Lu Chen arxiv

Long-horizon LLM agents require reinforcement learning methods that can assign credit to intermediate decisions under sparse and delayed rewards. Recent group-based methods such as GiGPO improve over GRPO by constructing step-level advantages at repeated anchor states. However, we show that such dense credit can be statistically unreliable: under limited rollouts, rare but lucky actions may receive overly large advantages, producing divergent anchor bias and late-stage training oscillation. We propose Evidence-Calibrated Policy Optimization (ECPO), a critic-free policy optimization algorithm that calibrates step-level credit before policy updates. ECPO combines Evidence-Calibrated Action Advantage, which groups rollouts by canonical actions and shrinks low-count estimates, with Variance-Gated Credit Weighting, which suppresses anchor states dominated by within-action noise. Experiments on ALFWorld and WebShop with Qwen2.5-1.5B/7B show that ECPO consistently outperforms strong baselines, improving GiGPO by +5.2/+7.3 success points on ALFWorld/WebShop with Qwen2.5-1.5B while adding only 0.1% additional advantage-computation overhead.

📄 PDF Abstract BibTeX arXiv:2606.05885

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning

2026-08-06 · Zi-Han Wang, Zhengxi Lu, Zhiyuan Yao, Jinyang Wu 외 hf

Reinforcement learning (RL) with verifiable rewards constructs trajectory-level advantage estimates, yet it often fails to credit the few pivotal decisions that determine outcomes in long-horizon, multi-turn agentic task…

Reinforcement Learning

PBSD: Privileged Bayesian Self-Distillation for Long-Horizon Credit Assignment

2026-06-08 · Yang Tian, Rui Wang, Xumeng Wen, Junjie Li 외 arxiv

Long-horizon agentic tasks pose a fundamental credit assignment challenge for outcome-base reinforcement learning: trajectory-level rewards verify final correctness but provide limited guidance on which intermediate reas…

Reinforcement Learning

MileGPO: Milestone Inference with Local Evidence for Graph-Based Policy Optimization of Long-Horizon LLM Agents

2026-08-20 · Bo Qian, Yuting Wu, Shuang Zeng, Huaiyu Wan 외 arxiv

Credit assignment is challenging in long-horizon agentic reinforcement learning, where supervision often comes only from final rewards. Existing methods refine trajectory-level signals into step-level credits through ste…

Reinforcement Learning

CreditDecoding: Accelerating Parallel Decoding in Diffusion Large Language Models with Trace Credit

2025-10-07 · Kangyu Wang, Zhiyun Jiang, Haibo Feng, Weijia Zhao 외 arxiv

Diffusion large language models (dLLMs) generate text through iterative denoising. In commonly adopted parallel decoding schemes, each step confirms only high-confidence positions while remasking the others. By analyzing…

Harness-G: A Graph-Structured Harness for Search Agents

2026-07-30 · Yanning Hou, Haoyuan Chen, Sihang Zhou, Xiaoshu Chen 외 arxiv

Reinforcement learning (RL) search agents commonly model retrieval as free-form natural-language query generation and optimize multi-turn interactions using final-answer rewards. Current studies mainly improve training w…

Reinforcement Learning