arXiv cs.CL· Baibei Ji, Xiaoyang Weng, Juntao Li, Zecheng Tang, Yihang Lou, Min Zhang·· 6 天前AI 评分23
MemReread：通过记忆引导重读增强智能体长上下文推理
MemReread: Enhancing Agentic Long-Context Reasoning via Memory-Guided Rereading
AI 导读
MemReread 提出一种基于流式阅读的智能体记忆方法，在最终记忆不足时触发问题分解与定向重读，绕过中间检索以恢复被过早丢弃的间接证据。其配套强化学习框架提升长度外推能力，并按任务复杂度自适应决定重读次数。MemReread-4B 在 2WikiMultiHopQA 上较 DeepSeek-V4-Pro 提升 1.6pp，同时保持相对上下文长度的线性时间复杂度。
正文 · 原文
Abstract:Long-context reasoning tasks have motivated agent-memory approaches that process document chunks sequentially while maintaining a dynamically updated memory, avoiding the quadratic complexity of standard attention. Existing methods often rely on retrieval to recover information discarded during memory overwriting, but remain vulnerable to evidence loss during memory formation and interference from invalid queries. To overcome these limitations, we propose MemReread. Built upon streaming reading, it circumvents intermediate retrieval, triggering question decomposition and targeted rereading when the final memory is insufficient. By recovering indirect evidence that may have been prematurely discarded, MemReread enables non-linear reasoning while preserving the logical flow of document comprehension. We further introduce a reinforcement learning framework that improves length extrapolation and adaptively determines the number of rereading passes according to task complexity, enabling flexible control of computational overhead. Our MemReread-4B outperforms strong baselines on long-context reasoning tasks (e.g., +1.6pp v.s. DeepSeek-V4-Pro on 2WikiMultiHopQA) while maintaining linear time complexity with respect to context length.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.10268 [cs.CL] |
| (or arXiv:2605.10268v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.10268 arXiv-issued DOI via DataCite |
Submission history
From: Baibei Ji [view email]
[v1]
Mon, 11 May 2026 09:30:59 UTC (556 KB)
[v2]
Fri, 9 Oct 2026 03:40:27 UTC (504 KB)
来源：arXiv cs.CL · arxiv.org