标题： 分布式检索增强生成 显示英文标题

标题： Distributed Retrieval-Augmented Generation

摘要： 随着大型语言模型（LLMs）在边缘设备上的应用越来越广泛， Retrieval-Augmented Generation (RAG) 作为一种通过整合外部知识来解决事实性缺陷和幻觉问题的解决方案，正变得越来越重要。然而，集中式 RAG 架构在数据隐私和可扩展性方面面临着重大挑战。例如，智能医疗保健服务通常依赖于收集敏感患者数据并建立一个集中式知识库，以提供更好的诊断和治疗建议，而隐私问题严重阻碍了这一过程。此外，维护一个全面且不断更新的知识库成本高昂，尤其是在应对区域性流行病和快速变异病毒的情况下。 为了解决这些挑战，本文介绍了一种名为分布式 Retrieval-Augmented Generation (DRAG) 的新框架，该框架通过消除对集中式知识库的需求并恢复数据控制权给所有者，从而提高了数据隐私。DRAG 结合了一种主题感知随机游走（TARW）算法，该算法利用 LLMs 提取查询主题，并在点对点网络中促进目标对等发现，从而实现在去中心化环境中的高效知识检索。 在三个不同数据集和 LLMs 上进行的广泛实验表明，使用 TARW 的 DRAG 在消息数量仅为泛洪一半的情况下实现了接近集中式 RAG 的性能。代码可在 https://github.com/xuchenhao001/DRAG 获取。 显示英文摘要

摘要： As large language models (LLMs) become increasingly adopted on edge devices, Retrieval-Augmented Generation (RAG) is gaining prominence as a solution to address factual deficiencies and hallucinations by integrating external knowledge. However, centralized RAG architectures face significant challenges in data privacy and scalability. For instance, smart healthcare services often rely on collecting sensitive patient data and building a centralized knowledge base to provide better diagnosis and treatment advice, while privacy concerns significantly impede this process. Besides, maintaining a comprehensive and continuously updated knowledge base is costly, particularly in response to regional epidemics and rapidly mutating viruses. To address these challenges, this paper introduces Distributed Retrieval-Augmented Generation (DRAG), a novel framework that improves data privacy by eliminating the need for a centralized knowledge base and restoring data control to owners. DRAG incorporates a Topic-Aware Random Walk (TARW) algorithm that leverages LLMs to extract query topics and facilitate targeted peer discovery within a peer-to-peer network, enabling efficient knowledge retrieval in decentralized environments. Extensive experiments across three diverse datasets and LLMs demonstrate that DRAG with TARW achieves near-centralized RAG performance by using half as many messages as flooding. The code is available at https://github.com/xuchenhao001/DRAG.