arXiv cs.AI· Luoxi Tang, Yuqiao Meng, Nilesh Auradkar, Muchao Ye, Dazheng Zhang, Zhaohan Xi·· 10 天前AI 评分31
如何理解并缓解 LLM 智能体的推理时记忆过度依赖：MEMTRIM 框架
Understanding and Mitigating Inference-Time Overreliance Using Agentic Memory
AI 导读
研究发现，LLM 智能体的 agentic memory 即使存储正确、检索得当，在查询与记忆仅部分重叠时仍会扭曲推理，造成"记忆过度依赖"。为此提出的 MEMTRIM 在写入时索引记忆证据、读取时控制复用，去除重复或冲突证据并保留有用信息，无需重训练，适用于基于嵌入和结构化记忆。实验显示 MEMTRIM 在多种模型和记忆设置下降低了记忆过度依赖，同时保留有用记忆的收益。
正文 · 原文
Abstract:Agentic memory allows LLM agents to reuse past experience, yet retrieved memories can also distort inference even when they are benign, correctly stored, and appropriately retrieved. We study this failure mode, which we call memory over-reliance. Across benchmarks and memory architectures, we find that memory is useful when past experience transfers to the current task, but can become misleading when only part of the evidence transfers. Failures are strongest under partial query-memory overlap, a pattern further confirmed by controlled experiments thatvary the amount of overlapping evidence. Motivated by this finding, we propose MEMTRIM, a plug-and-play framework that indexes memory evidence at write time and controls its reuse at read time. MEMTRIM removes repeated or conflicting evidence while preserving useful memory-specific information, requires no retraining, and applies to both embedding-based and structured memory this http URL show that MEMTRIM reduces memory overreliance while preserving the benefits of useful memory across models and memory settings.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.07311 [cs.AI] |
| (or arXiv:2610.07311v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.07311 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Luoxi Tang [view email]
[v1]
Mon, 5 Oct 2026 19:48:54 UTC (1,191 KB)
来源：arXiv cs.AI · arxiv.org