RAG / 检索增强生成 - arXivDaily 专题

2305.19912 2023-06-01 cs.IR 79%

Structure-Aware Language Model Pretraining Improves Dense Retrieval on Structured Data

Xinze Li, Zhenghao Liu, Chenyan Xiong, Shi Yu, Yu Gu, Zhiyuan Liu, Ge Yu

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

Comments Accepted by Findings of ACL 2023

2304.13654 2023-04-27 cs.IR 79%

A Personalized Dense Retrieval Framework for Unified Information Access

Hansi Zeng, Surya Kallumadi, Zaid Alibadi, Rodrigo Nogueira, Hamed Zamani

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

Comments Accepted to SIGIR 2023

2209.00179 2023-02-07 cs.IR 79%

Universal Vision-Language Dense Retrieval: Learning A Unified Representation Space for Multi-Modal Retrieval

Zhenghao Liu, Chenyan Xiong, Yuanhuiyi Lv, Zhiyuan Liu, Ge Yu

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

Comments Accepted by ICLR 2023

2205.03284 2022-10-25 cs.IR 79%

Dimension Reduction for Efficient Dense Retrieval via Conditional Autoencoder

Zhenghao Liu, Han Zhang, Chenyan Xiong, Zhiyuan Liu, Yu Gu, Xiaohua Li

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

Comments Accepted by EMNLP 2022

2206.03281 2022-06-08 cs.IR 79%

Unsupervised Context Aware Sentence Representation Pretraining for Multi-lingual Dense Retrieval

Ning Wu, Yaobo Liang, Houxing Ren, Linjun Shou, Nan Duan, Ming Gong, Daxin Jiang

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

2108.11480 2021-08-27 cs.IR 79%

On Approximate Nearest Neighbour Selection for Multi-Stage Dense Retrieval

Craig Macdonald, Nicola Tonellotto

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

2108.10341 2021-08-25 cs.IR 79%

Query Embedding Pruning for Dense Retrieval

Nicola Tonellotto, Craig Macdonald

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

2106.11251 2021-07-02 cs.IR 79%

Pseudo-Relevance Feedback for Multiple Representation Dense Retrieval

Xiao Wang, Craig Macdonald, Nicola Tonellotto, Iadh Ounis

专题命中向量检索：dense retrieval(title,abstract)；分类 cs.IR

Comments 10 pages

Journal ref Proceedings of ICTIR 2021

2608.16044 2026-08-18 cs.CR cs.CL cs.IR 新提交 79%

Coverage Is Not Containment: A Fundamental Limit of Admission-Time Defenses Against Coordinated Poisoning of Vector Retrieval

覆盖并非包含：针对向量检索协同投毒的准入时刻防御的根本局限

Prashant Kumar Pathak, Tarun Kumar Sharma

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.IR、cs.CL

AI总结该研究揭示摄入时刻防御无法抵御向量检索协同投毒，提出检索时刻检测器可捕获全部攻击，证明覆盖并非包含，稳健防御需转向查询需求。

Comments 10 pages, 9 figures. Preprint; under submission

2606.18497 2026-06-23 cs.CR cs.AI cs.DB 新提交 79%

Ghost Vectors: Soft-Deleted Embeddings Remain Reconstructible in HNSW Vector Databases

幽灵向量：HNSW向量数据库中软删除的嵌入仍然可重构

Chandranil Chakraborttii, Jackeline García Alvarado, Sitora Abdulofizova, Shivanshu Dwivedi

机构 * Trinity College, USA（美国三一学院）

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.AI、cs.DB

AI总结研究揭示HNSW向量数据库的软删除机制存在安全漏洞，被标记删除的向量仍可通过存储层恢复，并提出基于加密密钥轮换的防护方案。

Comments 13 pages, 5 figures, 12 tables. Prepared for submission

2606.19692 2026-06-19 cs.CR cs.DB cs.IR 新提交 79%

When Global Gating Is Enough: Admission-Time Hubness Control in Anisotropic Vector Retrieval Systems

当全局门控足够：各向异性向量检索系统中的准入时间枢纽性控制

Prashant Kumar Pathak, Tarun Kumar Sharma

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.IR、cs.DB

AI总结针对检索增强生成中向量枢纽性引发的投毒风险，提出准入时间控制方法，通过哨兵查询评分隔离枢纽文档，全局门控在多个数据集上达到高召回率和低误报率。

2602.11443 2026-02-13 cs.DB cs.IR 79%

Filtered Approximate Nearest Neighbor Search in Vector Databases: System Design and Performance Analysis

向量数据库中的过滤近邻搜索：系统设计与性能分析

Abylay Amanbayev, Brian Tsan, Tri Dang, Florin Rusu

专题命中向量检索：retrieval-augmented generation(abstract)；RAG(abstract)；vector search(abstract)；分类 cs.IR、cs.DB

AI总结本文研究了向量数据库中过滤近邻搜索的系统设计与性能，通过MoReVec数据集评估了不同引擎的过滤策略表现，揭示了算法适应性对索引性能的影响，并提出了GLS相关度量指标。

Comments The artifacts are available at: https://github.com/aabylay/ANN-benchmark-HQ

2605.17415 2026-05-25 cs.LG cs.AI cs.DB cs.IR 78%

IVF-TQ: Calibration-Free Streaming Vector Search via a Codebook-Free Residual Layer

IVF-TQ：通过无码本残差层实现无需校准的流式向量搜索

Tarun Sharma

机构 * Independent Researcher（独立研究者）

专题命中向量检索：vector search(title)；分类 cs.IR、cs.AI、cs.DB

AI总结本文提出IVF-TQ，一种基于数据无关残差压缩层的倒排文件索引，通过固定随机旋转和预计算Lloyd-Max标量量化器实现流式向量搜索的稳定性能，无需码本训练或逐数据集位预算调整。

2605.11523 2026-05-13 cs.DC 78%

NAVIS: Concurrent Search and Update with Low Position-Seeking Overhead in On-SSD Graph-Based Vector Search

NAVIS: 在SSD上实现低位置搜索开销的并发搜索与更新

Jaeyong Song, Hongsun Jang, Changmin Shin, Seongyeon Park, Yong Jae Ryoo, Seo Jin Park, Jinho Lee

专题命中向量检索：vector search(title,abstract)

AI总结 NAVIS通过优化布局、动态入口图和边列表缓存，降低位置搜索开销，提升大规模高维数据库的插入和搜索吞吐量，同时减少搜索延迟。

Comments 14 pages

2506.20584 2025-12-22 cs.LG 78%

The kernel of graph indices for vector search

图索引的核方法用于向量搜索

Mariano Tepper, Ted Willke

专题命中向量检索：vector search(title,abstract)

AI总结本文提出了一种基于核方法的图索引SVG，适用于度量和非度量向量空间，并引入SVG-L0实现稀疏性约束，提升向量搜索效率。

2512.02278 2025-12-03 cs.DC 78%

Fantasy: Efficient Large-scale Vector Search on GPU Clusters with GPUDirect Async

Fantasy: 在GPU集群上高效进行大规模向量搜索

Yi Liu, Chen Qian

专题命中向量检索：vector search(title,abstract)

AI总结 Fantasy通过在GPU集群中流水线向量搜索与数据传输，利用GPUDirect Async技术提高大规模向量搜索的效率和吞吐量。

2406.12385 2025-07-21 cs.AR 78%

Fast Graph Vector Search via Hardware Acceleration and Delayed-Synchronization Traversal

Wenqi Jiang, Hang Hu, Torsten Hoefler, Gustavo Alonso

专题命中向量检索：vector search(title,abstract)

Comments Accepted by VLDB'25

Journal ref Proceedings of the VLDB Endowment Volume 18, 2025

2507.06653 2025-07-10 cs.DC 78%

Towards Efficient and Scalable Distributed Vector Search with RDMA

Xiangyu Zhi, Meng Chen, Xiao Yan, Baotong Lu, Hui Li, Qianxi Zhang, Qi Chen, James Cheng

专题命中向量检索：vector search(title,abstract)

2502.16931 2025-02-25 cs.LG 78%

Machine learning and high dimensional vector search

Matthijs Douze

专题命中向量检索：vector search(title,abstract)

2102.09206 2021-09-17 cs.LG 78%

Less is More: Pre-train a Strong Text Encoder for Dense Retrieval Using a Weak Decoder

Shuqi Lu, Di He, Chenyan Xiong, Guolin Ke, Waleed Malik, Zhicheng Dou, Paul Bennett, Tieyan Liu, Arnold Overwijk

专题命中向量检索：dense retrieval(title,abstract)

1809.04067 2018-09-13 cs.CV cs.LG cs.PF 78%

Zoom: SSD-based Vector Search for Optimizing Accuracy, Latency and Memory

Minjia Zhang, Yuxiong He

专题命中向量检索：vector search(title,abstract)

2607.29173 2026-08-20 cs.DB 版本更新 77%

MERIT: Efficient In-Place Deletion for Dynamic Graph-Based Approximate Nearest Neighbor Indexes

MERIT：基于动态图的近似最近邻索引的高效原位删除方法

Zekai Wu, Jiabao Jin, Peng Cheng, Wangze Ni, Haoyang Li, Lei Chen, Junjie Yao, Jingkuan Song, Heng Tao Shen

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.DB

AI总结本文针对动态图近似最近邻索引的原位删除难题，提出MERIT框架，通过三项核心技术实现高效删除，集成于两种主流索引，实验显示其删除速度远优于SOTA方法且召回率稳定。

Comments 14 pages

2605.24930 2026-08-19 cs.CL 版本更新 77%

H$^{2}$MT: Semantic Hierarchy-Aware Hierarchical Memory Transformer

H$^{2}$MT: 语义层次感知的层次记忆Transformer

Maryam Haghifam, Zifan He, Jason Cong, Yizhou Sun

机构 * University of California, Los Angeles（加州大学洛杉矶分校）

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.CL

AI总结提出H$^{2}$MT模型，通过离线构建语义层次结构并利用自底向上的后序聚合计算记忆嵌入，在推理时实现从粗到细的查询路由，从而在长上下文推理中实现质量与效率的权衡。

2608.12812 2026-08-14 cs.IR 新提交 77%

A Comprehensive Empirical Evaluation of Vector Database Systems for Approximate Nearest Neighbor Search: Performance, Quality, and Resource Trade-offs

用于近似最近邻搜索的向量数据库系统的综合实证评估：性能、质量与资源权衡

Ashen Rashmiks, Tiroshan Madushanka

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.IR

AI总结本文对7款主流向量数据库系统开展综合实证评估，测量多维度指标，推导系统选择指南并发布开源基准框架，为AI应用选型提供依据。

2505.17810 2026-08-05 cs.LG cs.IR 版本更新 77%

VIBE: Vector Index Benchmark for Embeddings

VIBE：用于嵌入的向量索引基准测试

Elias Jääsaari, Ville Hyvönen, Matteo Ceccarello, Teemu Roos, Martin Aumüller

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.IR

AI总结本研究针对现有ANN基准数据集无法代表现代应用的问题，推出开源VIBE框架，含分布内外数据集，评估22个向量索引，为ANN算法提供最新基准测试。

Comments The 2nd Workshop on Vector Databases (VecDB@VLDB2026)

2607.04630 2026-07-07 cs.DB 新提交 77%

Exploiting Structural Properties for Efficient Constraint-Aware HNSW Hyperparameter Tuning

利用结构特性进行高效的约束感知HNSW超参数调优

Geon Choi, Hoeun Lee, Jaeyoung Do

专题命中向量检索：retrieval-augmented generation(abstract)；RAG(abstract)；vector search(abstract)；分类 cs.DB

AI总结从系统角度研究HNSW超参数调优，发现其配置空间有结构规律，提出CHAT框架，利用特定结构进行确定性、高效搜索，在多数据集和引擎上表现出色，优于基线。

Comments This paper has been accepted to ACM SIGMOD 2027

2606.00734 2026-06-02 cs.DB 77%

EMA: Approximate Nearest Neighbor Search with General Attribute Filtering and Dynamic Updates

EMA：支持通用属性过滤和动态更新的近似最近邻搜索

Mocheng Li, Baotong Lu, James Cheng, Chenhao Ma

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.DB

AI总结提出EMA算法，通过引入附着在图边上的紧凑摘要Marker，实现支持混合数值和类别属性多谓词查询的过滤近似最近邻搜索，并支持高效动态更新，在多种工作负载下速度提升1.68至12.25倍。

Comments 13 pages, 10 figures, Submitted to PVLDB Research Track

2605.21952 2026-05-22 cs.AR cs.DB cs.DC 77%

NasZip: Software and Hardware Co-Design to Accelerate Approximate Nearest Neighbor Search with DIMM-Based Near-Data Processing

NasZip: 软件和硬件协同设计以加速基于DIMM的近数据处理的近似最近邻搜索

Cheng Zou, Shuo Yang, Chen Nie, Yu Zou, Yu He, Chao Jiang, Limin Xiao, Weifeng Zhang, Zhezhi He

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.DB

AI总结本文提出NasZip框架，通过结合近数据处理和基于统计的主成分分析的特征级早退技术，优化高维向量距离计算，从而在保持准确性的同时提升近似最近邻搜索的效率。

Comments 17 pages, accepted by Proceedings of the 53rd Annual International Symposium on Computer Architecture (ISCA-26)

2605.15184 2026-05-15 cs.CL 77%

Is Grep All You Need? How Agent Harnesses Reshape Agentic Search

Grep 是否足够？代理如何重塑搜索流程

Sahil Sen, Akhil Kasturi, Elias Lumer, Anmol Gulati, Vamse Kumar Subbiah

机构 * PricewaterhouseCoopers（普华永道）

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.CL

AI总结本文通过两个实验比较了grep与向量检索在代理搜索中的效果，发现grep在准确性上优于向量检索，但整体表现仍受代理架构和工具调用方式影响。

2605.13764 2026-05-14 cs.CR cs.IR cs.LG 77%

VectorSmuggle: Steganographic Exfiltration in Embedding Stores and a Cryptographic Provenance Defense

VectorSmuggle: 嵌入式存储中的隐写术外泄及加密溯源防御

Jascha Wanger

机构 * ThirdKey / Tarnover, LLC（ThirdKey / 塔诺弗公司）

专题命中向量检索：RAG(abstract,abstract_cn)；retrieval-augmented generation(abstract)；分类 cs.IR

AI总结本文研究了在嵌入式存储中通过简单扰动隐藏数据的隐写术攻击，提出VectorPin协议通过加密签名保障嵌入数据完整性，解决了嵌入式存储的溯源和完整性问题。

Comments 47 pages, 3 figures. Reference implementations: https://github.com/jaschadub/VectorSmuggle and https://github.com/jaschadub/VectorPin

1. 向量检索 348 篇

Structure-Aware Language Model Pretraining Improves Dense Retrieval on Structured Data

A Personalized Dense Retrieval Framework for Unified Information Access

Universal Vision-Language Dense Retrieval: Learning A Unified Representation Space for Multi-Modal Retrieval

Dimension Reduction for Efficient Dense Retrieval via Conditional Autoencoder

Unsupervised Context Aware Sentence Representation Pretraining for Multi-lingual Dense Retrieval

On Approximate Nearest Neighbour Selection for Multi-Stage Dense Retrieval

Query Embedding Pruning for Dense Retrieval

Pseudo-Relevance Feedback for Multiple Representation Dense Retrieval

Coverage Is Not Containment: A Fundamental Limit of Admission-Time Defenses Against Coordinated Poisoning of Vector Retrieval

Ghost Vectors: Soft-Deleted Embeddings Remain Reconstructible in HNSW Vector Databases

When Global Gating Is Enough: Admission-Time Hubness Control in Anisotropic Vector Retrieval Systems

Filtered Approximate Nearest Neighbor Search in Vector Databases: System Design and Performance Analysis

IVF-TQ: Calibration-Free Streaming Vector Search via a Codebook-Free Residual Layer

NAVIS: Concurrent Search and Update with Low Position-Seeking Overhead in On-SSD Graph-Based Vector Search

The kernel of graph indices for vector search

Fantasy: Efficient Large-scale Vector Search on GPU Clusters with GPUDirect Async

Fast Graph Vector Search via Hardware Acceleration and Delayed-Synchronization Traversal

Towards Efficient and Scalable Distributed Vector Search with RDMA

Machine learning and high dimensional vector search

Less is More: Pre-train a Strong Text Encoder for Dense Retrieval Using a Weak Decoder

Zoom: SSD-based Vector Search for Optimizing Accuracy, Latency and Memory

MERIT: Efficient In-Place Deletion for Dynamic Graph-Based Approximate Nearest Neighbor Indexes

H$^{2}$MT: Semantic Hierarchy-Aware Hierarchical Memory Transformer

A Comprehensive Empirical Evaluation of Vector Database Systems for Approximate Nearest Neighbor Search: Performance, Quality, and Resource Trade-offs

VIBE: Vector Index Benchmark for Embeddings

Exploiting Structural Properties for Efficient Constraint-Aware HNSW Hyperparameter Tuning

EMA: Approximate Nearest Neighbor Search with General Attribute Filtering and Dynamic Updates

NasZip: Software and Hardware Co-Design to Accelerate Approximate Nearest Neighbor Search with DIMM-Based Near-Data Processing

Is Grep All You Need? How Agent Harnesses Reshape Agentic Search

VectorSmuggle: Steganographic Exfiltration in Embedding Stores and a Cryptographic Provenance Defense