研究库 论文知识库

内容库 / 论文

arXiv:2610.13314 · RAG 检索增强

NEAR: Neighborhood-Aware Residuals for Memory Retrieval

NEAR：面向记忆检索的邻域感知残差

NEAR: Neighborhood-Aware Residuals for Memory Retrieval

类型：arxiv
标识：2610.13314
链接：http://arxiv.org/abs/2610.13314v1
主分类：rag
形态：method
TLDR：Memory retrieval aims to identify evidence from stored records to answer a query. Stored records with different factual details may share topics, entities, or events. Such shared contents would make records without the required evidence more likely to receive high similarity scores and outrank the correct evidence. In this paper, we use the information shared by these records to construct local references for query-memory matching. We propose NEighborhood-Aware Residuals (NEAR), a training-free framework that uses these references to construct residual representations that complement the origi
待LLM分类：否
标题中文：NEAR：面向记忆检索的邻域感知残差
TLDR中文：记忆检索旨在从存储记录中识别证据以回答查询。事实细节不同的存储记录可能共享主题、实体或事件。这种共享内容会使缺乏所需证据的记录更容易获得高相似度分数，从而排在正确证据之前。本文利用这些记录共享的信息，为查询-记忆匹配构建局部参考。我们提出邻域感知残差（NEAR），一个免训练框架，利用这些参考构建与原始……互补的残差表示
来源文件：
/inbox/tom/_candidates/2026-10-12-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记 Cool Papers · cs.CL (papers.cool) · RSS 摘要

笔记 DataSpace · 短审稿（2026-08-10 09:50 · flyP 精读棒）

笔记 Tom 文献雷达 · AI Agent / RAG / 长上下文 · 2026-07-11 下午版

笔记 Tom 文献雷达 · Agent · RAG · 长上下文 · 2026-08-16 08:40 UTC

文档 Tom 评 flyP · 2026-09-28 早棒

笔记 ReMemR1: Look Back to Reason Forward (Revisitable Memory for Long-Context LLM Agents)

arXiv

2610.13314

TLDR

记忆检索旨在从存储记录中识别证据以回答查询。事实细节不同的存储记录可能共享主题、实体或事件。这种共享内容会使缺乏所需证据的记录更容易获得高相似度分数，从而排在正确证据之前。本文利用这些记录共享的信息，为查询-记忆匹配构建局部参考。我们提出邻域感知残差（NEAR），一个免训练框架，利用这些参考构建与原始……互补的残差表示Memory retrieval aims to identify evidence from stored records to answer a query. Stored records with different factual details may share topics, entities, or events. Such shared contents would make records without the required evidence more likely to receive high similarity scores and outrank the correct evidence. In this paper, we use the information shared by these records to construct local references for query-memory matching. We propose NEighborhood-Aware Residuals (NEAR), a training-free framework that uses these references to construct residual representations that complement the origi

arXiv ↗ 中文全文