arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.14374cs.CRcs.AIcs.CL
收藏

差分隐私增强的检索增强生成

Differentially Private Retrieval-Augmented Generation

University of Southern California(南加州大学)

Tingting Tang, James Flemings, Yongqin Wang, Murali Annavaram

AI总结:

DP-KSA 是一种整合差分隐私的检索增强生成算法,通过关键词增强减少隐私风险并保持实用性。

摘要:

检索增强生成(RAG)是一种广泛使用的框架,用于在特定领域任务中通过从数据库中检索相关文档来减少大语言模型(LLMs)的幻觉。然而,当数据库包含敏感语料,如医疗记录或法律文件时,RAG 通过其输出可能暴露私人信息,从而带来严重的隐私风险。先前的研究表明,人们可以实际地构造对抗性提示,迫使 LLM 重新吐出增强的上下文。一个有希望的方向是将差分隐私(DP),一种提供强形式保证的隐私概念,整合到 RAG 系统中。然而,简单地将 DP 机制应用于现有系统往往会导致显著的实用性下降。特别是对于 RAG 系统,DP 可能会降低增强上下文的有用性,从而增加 LLM 的幻觉风险。受这些挑战的启发,我们提出了 DP-KSA,一种新颖的隐私保护 RAG 算法,它通过提出-测试-发布范式整合 DP。DP-KSA 基于一个关键观察,即大多数问答(QA)查询可以用少量关键词充分回答。因此,DP-KSA 首先获得一组相关的上下文,每个都会用于生成 LLM 的响应。我们利用这些响应来获得在差分隐私方式下最频繁的关键词。最后,这些关键词被增强到最终输出的提示中。这种方法有效地压缩了语义空间,同时保持了实用性和隐私。我们正式证明 DP-KSA 对 RAG 数据库生成的输出提供了形式上的 DP 保证。我们在两个 QA 评估基准上评估了 DP-KSA,使用三个指令调优的 LLMs,我们的实证结果表明 DP-KSA 实现了强大的隐私-实用性权衡。

英文摘要:

Retrieval-augmented generation (RAG) is a widely used framework for reducing hallucinations in large language models (LLMs) on domain-specific tasks by retrieving relevant documents from a database to support accurate responses. However, when the database contains sensitive corpora, such as medical records or legal documents, RAG poses serious privacy risks by potentially exposing private information through its outputs. Prior work has demonstrated that one can practically craft adversarial prompts that force an LLM to regurgitate the augmented contexts. A promising direction is to integrate differential privacy (DP), a privacy notion that offers strong formal guarantees, into RAG systems. However, naively applying DP mechanisms into existing systems often leads to significant utility degradation. Particularly for RAG systems, DP can reduce the usefulness of the augmented contexts leading to increase risk of hallucination from the LLMs. Motivated by these challenges, we present DP-KSA, a novel privacy-preserving RAG algorithm that integrates DP using the propose-test-release paradigm. DP-KSA follows from a key observation that most question-answering (QA) queries can be sufficiently answered with a few keywords. Hence, DP-KSA first obtains an ensemble of relevant contexts, each of which will be used to generate a response from an LLM. We utilize these responses to obtain the most frequent keywords in a differentially private manner. Lastly, the keywords are augmented into the prompt for the final output. This approach effectively compresses the semantic space while preserving both utility and privacy. We formally show that DP-KSA provides formal DP guarantees on the generated output with respect to the RAG database. We evaluate DP-KSA on two QA benchmarks using three instruction-tuned LLMs, and our empirical results demonstrate that DP-KSA achieves a strong privacy-utility tradeoff.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑