发表机构
Binghamton University, State University of New York; University of Iowa; Penn Medicine, University of Pennsylvania（纽约州立大学宾汉姆顿分校; 爱荷华大学; 宾夕法尼亚大学佩雷尔曼医学院）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文研究LLM智能体记忆导致的推理过度依赖问题，提出即插即用框架MEMTRIM，通过写入时索引和读取时控制重用，减少过度依赖并保留记忆益处。
AI 中文摘要
智能体记忆使大型语言模型（LLM）智能体能够复用过去的经验，然而，即使检索到的记忆是无害的、正确存储且恰当检索的，它们也可能扭曲推理过程。我们研究了这一失败模式，称之为记忆过度依赖。在多个基准测试和记忆架构中，我们发现，当过去的经验能够迁移到当前任务时，记忆是有用的，但当只有部分证据迁移时，记忆可能产生误导。在查询与记忆部分重叠的情况下，失败最为严重，这一模式通过改变重叠证据数量的受控实验得到了进一步证实。基于这一发现，我们提出了MEMTRIM，一个即插即用的框架，在写入时索引记忆证据，并在读取时控制其重用。MEMTRIM移除重复或冲突的证据，同时保留有用的记忆特有信息，无需重新训练，适用于基于嵌入的和结构化的记忆。我们证明，MEMTRIM在多种模型和记忆设置下减少了记忆过度依赖，同时保留了有用记忆的益处。
英文摘要
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 systems.Experiments show that MEMTRIM reduces memory overreliance while preserving the benefits of useful memory across models and memory settings.
