arXiv：cs.LG（机器学习，全量分类）· Hyeonjeong Ha, Jeonghwan Kim, Cheng Qian, Jiayu Liu, William M. Campbell, Yue Wu, Yuji Zhang, Kathleen McKeown, Dilek Hakkani-Tur, Heng Ji·· 15 小时前AI 评分43
MemGuard：防止长期记忆增强大语言模型中的记忆污染
MemGuard: Preventing Memory Contamination in Long-Term Memory-Augmented Large Language Models
AI 导读
MemGuard 是一个类型感知的记忆框架，通过为每条记忆在写入时分配明确的功能角色、维护类型隔离记忆间的关系，并按需选择性组合证据，防止异构记忆污染。在幻觉与长程对话基准上，MemGuard 将记忆可靠性最高提升 28.27%，同时检索的 memory token 比先前方法最多减少 5.8 倍。
正文
Abstract:Memory-augmented large language models extend reasoning beyond a fixed context window by maintaining long-term memory across interactions. However, existing memory systems often collapse stable user facts, episodic events, and behavioral rules into a shared space, allowing functionally distinct memories to be retrieved and used as interchangeable evidence. We identify this failure mode as heterogeneous memory contamination, where context-specific events become overgeneralized claims, or semantically relevant but functionally incompatible memories mislead generation. To this end, we introduce MemGuard, a type-aware memory framework that preserves functional memory boundaries during memory construction and retrieval. It assigns each memory an explicit functional role at write time, maintains relations across type-isolated memories, and selectively composes evidence only from necessary memory types, reducing contamination from irrelevant or functionally incompatible evidence. Across hallucination and long-horizon conversation benchmarks, MemGuard improves memory reliability by up to 28.27% while retrieving up to 5.8x fewer memory tokens than prior methods. These results suggest that reliable long-term reasoning depends on principled organization and selective use of heterogeneous memory.
| Comments: | EMNLP 2026 |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.28009 [cs.CL] |
| (or arXiv:2605.28009v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.28009 arXiv-issued DOI via DataCite |
Submission history
From: Hyeonjeong Ha [view email]
[v1]
Wed, 27 May 2026 06:04:19 UTC (853 KB)
[v2]
Wed, 30 Sep 2026 20:01:49 UTC (857 KB)
来源：arXiv：cs.LG（机器学习，全量分类） · arxiv.org