从经验中学习状态化预测知识

AI 摘要

该论文提出状态知识学习（SKL）方法，使大语言模型代理从轨迹级反思转向维护显式的、基于状态的预测性知识。通过自蒸馏（SKL-SD）和强化学习（SKL-RL）两种算法，训练代理自主提取状态基础预测知识并用于决策。在WebShop、ScienceWorld和ChessPuzzles等交互环境中的实验表明，该方法显著优于当前基于反思的训练范式。

正文节选

Computer Science > Computation and Language Title:Learning Stateful Predictive Knowledge From Experience View PDF HTML (experimental) Abstract:As large language model (LLM) agents increasingly learn from experience, they primarily rely on trajectory-level reflection to extract insights. Viewed through the lens of predictive knowledge, we argue that this approach operates on episodic hindsight rather than predictive foresight, yielding brittle, path-dependent heuristics. To address th

发布时间：2026-08-03 12:00

抓取时间：2026-08-03 15:26

来源机构：arXiv