arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.14038cs.AIcs.LG
收藏

选择如何记忆:面向LLM代理的自适应记忆结构

Choosing How to Remember: Adaptive Memory Structures for LLM Agents

University of Technology Sydney(悉尼大学);University of Texas at Austin(德克萨斯大学奥斯汀分校);University of California, Los Angeles(加州大学洛杉矶分校);The University of Melbourne(墨尔本大学)

Mingfei Lu, Mengjia Wu, Feng Liu, Jiawei Xu, Weikai Li, Haoyang Wang, Zhengdong Hu, Ying Ding, Yizhou Sun, Jie Lu, Yi Zhang

AI总结:

本文提出FluxMem框架,通过自适应记忆结构选择和三级记忆层次结构,提升LLM代理在长时间交互中的表现,实验显示平均提升9.18%和6.14%。

摘要:

记忆对于使基于大语言模型(LLM)的代理在长时间跨度的交互中保持一致行为至关重要。然而,现有的代理记忆系统存在两个关键缺陷:它们依赖于一种通用记忆结构,并且不将记忆结构的选择建模为上下文自适应决策,限制了它们处理异构交互模式的能力,导致性能低下。我们提出了一个统一框架FluxMem,使LLM代理能够实现自适应的记忆组织。我们的框架为代理配备了多种互补的记忆结构。它明确学习根据交互层面的特征在这些结构之间进行选择,使用来自下游响应质量和记忆利用的离线监督。为了支持稳健的长时间记忆演化,我们进一步引入了三级记忆层次结构和基于Beta混合模型的概率门控,用于分布感知的记忆融合,取代了脆弱的相似度阈值。在两个长时间跨度基准测试PERSONAMEM和LoCoMo上的实验表明,我们的方法实现了平均提升9.18%和6.14%。

英文摘要:

Memory is critical for enabling large language model (LLM) based agents to maintain coherent behavior over long-horizon interactions. However, existing agent memory systems suffer from two key gaps: they rely on a one-size-fits-all memory structure and do not model memory structure selection as a context-adaptive decision, limiting their ability to handle heterogeneous interaction patterns and resulting in suboptimal performance. We propose a unified framework, FluxMem, that enables adaptive memory organization for LLM agents. Our framework equips agents with multiple complementary memory structures. It explicitly learns to select among these structures based on interaction-level features, using offline supervision derived from downstream response quality and memory utilization. To support robust long-horizon memory evolution, we further introduce a three-level memory hierarchy and a Beta Mixture Model-based probabilistic gate for distribution-aware memory fusion, replacing brittle similarity thresholds. Experiments on two long-horizon benchmarks, PERSONAMEM and LoCoMo, demonstrate that our method achieves average improvements of 9.18% and 6.14%.

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。

↑