Privacy-Preserving Reasoning with Knowledge-Distilled Parametric Retrieval Augmented Generation
This addresses privacy concerns for users of cloud-based RAG systems, though it is an incremental improvement over existing parametric RAG methods.
The paper tackles the problem of privacy leakage in retrieval-augmented generation (RAG) systems by proposing DistilledPRAG, a method that encodes documents as LoRA parameters without exposing raw content, achieving higher accuracy and better generalization on out-of-distribution data compared to baselines.
The current RAG system requires uploading plaintext documents to the cloud, risking private data leakage. Parametric RAG (PRAG) addresses this by encoding documents as LoRA within LLMs, enabling reasoning without exposing raw content. However, it still faces two issues: (1) PRAG demands synthesizing QA pairs and fine-tuning LLM for each individual document to create its corresponding LoRA, leading to unacceptable inference latency. (2) The performance of PRAG relies solely on synthetic QA data, lacking internal alignment with standard RAG, resulting in poor generalization on out-of-distribution(OOD) inputs. Therefore, achieving high-efficiency parameterization while maintaining RAG-level performance remains a critical challenge for privacy-preserving reasoning. In this paper, we propose DistilledPRAG, a generalizable knowledge-distilled parametric RAG model aligned with standard RAG in document structure and parameter activation. We first synthesize QA pairs from single and multi-documents to enhance cross-document reasoning. Then, we mask the plaintext documents with a special token and translate them to LoRA via a parameter generator, maintaining the standard RAG document structure. Finally, guided by synthetic QA data, we train the parameter generator to match standard RAG's hidden states and output logits, enabling RAG-style reasoning without original documents. Experiments on four QA datasets show that DistilledPRAG outperforms baselines in accuracy and generalizes well on OOD data.