매일 아침, 어제의 AI를 한 통으로 정리해 보내드립니다메일로 받아보기

METAL LAB

ERSkill: Evolving for Skill-Guided Adaptive Memory Retrieval

arXiv:2608.127202026-08-14

arXiv:2608.12720v1 Announce Type: new Abstract: While Large Language Model (LLM) agents increasingly rely on long-term memory for persistent interactions, the retrieval mechanisms governing this memory are rarely treated as evolvable components. This static approach limits performance on heterogeneous memory queries, which often demand diverse evidence construction strategies. To address this, we introduce \textbf{ERSkill}, a retrieval-centric framework for self-evolving, skill-guided memory acc

저자 · Haolong Chen, Liang Zhang, Zhuo Li, Lei Xue, Guanrxu Zhu

arXiv에서 원문 보기