arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.19951cs.CLcs.AIcs.LG
收藏

AtomSearch:基于证据的智能体技能优化树搜索

AtomSearch: Evidence-Grounded Tree Search for Agent Skill Optimization

Security Technology Inc.(360安全科技公司)

Jia Tong, Senyou Deng, Min Peng, Jianfeng Pan

AI总结:

AtomSearch是一种基于证据的树搜索方法,通过对比原子编辑、技能状态树搜索和结果记忆优化智能体技能,在四个任务上优于SkillOpt,在LiveMathematicianBench上获9.68个百分点增益。

摘要:

智能体技能通过可编辑的外部指令适配冻结模型。同时修改多条规则会使变更难以检查,而仅通过即时分数判断每个局部编辑可能会丢弃有用的中间状态。我们引入AtomSearch,该方法在保持修订操作局部性的同时,利用后续技能状态指导搜索。对比原子编辑将每个原始规则编辑与失败和成功的执行、一个行为假设以及预期范围关联起来。技能状态树搜索(STS)评估短编辑组合,并选择性地将其保留为可重新访问和扩展的可执行状态。结果记忆为后续提议提供观察到的修复和回归统计数据。我们在四个任务和五个目标模型上评估AtomSearch。使用deepseek-v4-flash时,它在四个任务中的三个上优于SkillOpt,包括在LiveMathematicianBench上获得9.68个百分点的增益,且SkillOpt的最终技能包含2.3至19.7倍的字符数。在67%的已完成搜索中,最佳验证状态经过了在其深度并非最高得分的祖先。这些结果表明,局部规则编辑如何支持有用修订路径的探索并产生紧凑技能,同时使用额外的离线评估来探索替代状态。

英文摘要:

Agent skills adapt frozen models through editable external instructions. Revising several rules at once makes changes difficult to inspect, while judging each local edit only by its immediate score can discard useful intermediate states. We introduce AtomSearch, which keeps revision actions local while using subsequent skill states to guide search. Contrastive atomic editing links each primitive rule edit to failed and successful executions, a behavioral hypothesis, and an intended scope. Skill-State Tree Search (STS) evaluates short edit compositions and selectively retains them as executable states that can be revisited and extended. Outcome memory supplies later proposals with observed repair and regression statistics. We evaluate AtomSearch across four tasks and five target models. With deepseek-v4-flash, it exceeds SkillOpt on three of four tasks, including a 9.68-percentage-point gain on LiveMathematicianBench, and SkillOpt's final skills contain 2.3-19.7 times as many characters. In 67% of completed searches, the best validation state passes through an ancestor that was not highest-scoring at its depth. These results show how local rule edits can support the exploration of useful revision paths and yield compact skills, with additional offline evaluations used to explore alternative states.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑