Skip to content
NewA calmer, faster ZeroHour
Researchinfo

Private Retrieval Augmented Generation via Random Projection

arXiv cs.CRby Dixi Yao ()1 min read4 · 2 reads
Read at arxiv.org
AI summary · grok-4.7Summaries by Vani AI

Random projection preserves RAG similarity and matches DP privacy with better retrieval utility.

The paper applies the same random projection to datastore embeddings and query embeddings so semantic similarity is preserved while extraction of sensitive retrieval data is reduced. Across multiple RAG architectures and datasets, the approach outperformed prior differential-privacy methods on retrieval and generation quality. Empirical privacy under strong attack models was comparable to those DP baselines. The authors propose random projection as a practical baseline for privacy-preserving RAG.

Key facts
  • Same random projection is applied to datastore and query embeddings.
  • Method provably preserves semantic similarity between queries and items.
  • Retrieval and generation beat differential-privacy baselines.
  • Empirical privacy under strong extraction attacks is comparable to DP.
Analyze this story with Vani AIImpact, who is affected and what defenders should do, in a new chat
Tags
ragprivacyrandom-projectiondifferential-privacydata-extractionembeddingsllm

From the original article

1 min read at arxiv.org

Retrieval-Augmented Generation (RAG) enhances the capabilities of large language models (LLMs) by querying external structured knowledge. However, it can also introduce privacy risks by leaking sensitive information from the retrieval datastore. We propose a simple method to preserve datastore privacy in RAG systems via random projection.

arXiv cs.CR

By Dixi Yao · arxiv.org

Read the full article at arXiv cs.CR

A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.