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 2602.13933cs.AI
收藏

HyMem:混合内存架构与动态检索调度

HyMem: Hybrid Memory Architecture with Dynamic Retrieval Scheduling

ZJU-UIUC Institute(浙大-伊利诺伊大学联合研究院);Ant Group(蚂蚁集团)

Xiaochen Zhao, Kaikai Wang, Xiaowen Zhang, Chen Yao, Aili Wang

AI总结:

HyMem通过多粒度内存表示实现动态按需调度,结合轻量级模块和LLM深度模块,有效平衡效率与性能,在长时记忆管理中取得新突破。

摘要:

大型语言模型(LLM)代理在短文本情境中表现优异,但在扩展对话中因内存管理低效而表现不佳。现有方法面临效率与效果的权衡:内存压缩可能丢失复杂推理所需的关键细节,而保留原始文本会为简单查询带来不必要的计算开销。核心问题在于单一内存表示和静态检索机制的局限性,无法模拟人类灵活主动的记忆调度能力,难以适应多样的问题场景。受认知经济原则启发,我们提出HyMem,一种混合内存架构,通过多粒度内存表示实现动态按需调度。HyMem采用双粒度存储方案配合动态双层检索系统:轻量模块构建摘要级上下文以实现高效响应生成,而LLM深度模块仅在复杂查询中被选择性激活,并通过反思机制进行迭代推理优化。实验表明,HyMem在LOCOMO和LongMemEval基准测试中表现强劲,优于全上下文方法,同时将计算成本降低92.6%,在长时记忆管理的效率与性能之间建立了最先进的平衡。

英文摘要:

Large language model (LLM) agents demonstrate strong performance in short-text contexts but often underperform in extended dialogues due to inefficient memory management. Existing approaches face a fundamental trade-off between efficiency and effectiveness: memory compression risks losing critical details required for complex reasoning, while retaining raw text introduces unnecessary computational overhead for simple queries. The crux lies in the limitations of monolithic memory representations and static retrieval mechanisms, which fail to emulate the flexible and proactive memory scheduling capabilities observed in humans, thus struggling to adapt to diverse problem scenarios. Inspired by the principle of cognitive economy, we propose HyMem, a hybrid memory architecture that enables dynamic on-demand scheduling through multi-granular memory representations. HyMem adopts a dual-granular storage scheme paired with a dynamic two-tier retrieval system: a lightweight module constructs summary-level context for efficient response generation, while an LLM-based deep module is selectively activated only for complex queries, augmented by a reflection mechanism for iterative reasoning refinement. Experiments show that HyMem achieves strong performance on both the LOCOMO and LongMemEval benchmarks, outperforming full-context while reducing computational cost by 92.6\%, establishing a state-of-the-art balance between efficiency and performance in long-term memory management.

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。

↑