[论文] SPEM: Structure-Aware Procedural Memory for Efficient Skill Use in LLM...

论文概要

研究领域: ML 作者: Ran Zhou, Zehui Kong, Fanyu Lin, Jiaheng Liu, Zhen Zhang, Yuxiang Ren 发布时间: 2026-10-22 arXiv: 2610.26995

中文摘要

基于LLM的智能体越来越依赖可复用的技能。然而，技能库的规模化使得高效且准确地使用技能变得困难。全量加载技能会产生高昂的token和运行时开销。相比之下，扁平的语义检索将技能视为独立的文本条目，忽略了依赖关系、执行顺序、状态转移和数据流约束。我们提出SPEM，一个轻量级的结构感知过程记忆框架，用于LLM智能体的高效技能使用。SPEM维护原子技能记忆（单个技能）和技能DAG记忆（从成功执行中派生的可复用组合）。对于新任务，SPEM在可用时复用高置信度的技能DAG模板；否则，检索原子技能并调用Skill-GNN进行状态和结构感知的技能选择。成功的执行被写回为新的技能DAG，使过程记忆能够随时间积累。在ALFWorld和SkillsBench上的实验表明，SPEM在整体效能-效率权衡上表现最佳，达到87.1%的ALFWorld成功率和38.38%的SkillsBench奖励，同时降低了token和运行时成本。

原文摘要

LLM-based agents increasingly rely on reusable skills. However, scaling skill libraries makes efficient and accurate skill use difficult. Full skill loading incurs high token and runtime overhead. In contrast, flat semantic retrieval treats skills as independent text entries, ignoring dependencies, execution order, state transitions, and data flow constraints. We propose SPEM, a lightweight structure-aware procedural memory framework for efficient skill use in LLM-based agents. SPEM maintains Atomic Skill Memory for individual skills and Skill-DAG Memory for reusable compositions derived from successful executions. For a new task, SPEM reuses high-confidence Skill-DAG templates when available. Otherwise, it retrieves atomic skills and invokes Skill-GNN for state- and structure-aware skill ...

#论文 #arXiv #ML #小凯