Privacy-Preserving Retrieval-Augmented Generation with Differential Privacy

Tatsuki Koga ∗ , Ruihan Wu ∗, Zhiyuan Zhang, Kamalika Chaudhuri · 2024

Read Privacy-Preserving Retrieval-Augmented Generation with Differential Privacy by Tatsuki Koga ∗ , Ruihan Wu ∗, Zhiyuan Zhang, Kamalika Chaudhuri (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-04697.txt.

Description

Tatsuki Koga, Ruihan Wu, Zhiyuan Zhang, and Kamalika Chaudhuri present privacy-preserving retrieval-augmented generation with differential privacy, in a research paper from 2024. Retrieval-augmented systems feed private documents to a language model at answer time. Without guards, answers can leak sensitive facts from those sources, which blocks use in fields such as health and law.

Their algorithm spends the privacy budget only on tokens that need the private data. All other tokens come from the plain, non-private model, which saves budget for accuracy where it counts. Tests across models and data sets beat the no-retrieval baseline at a moderate privacy level. Long, accurate answers become feasible under a formal privacy guarantee.

Catalog record

CREATORTatsuki Koga ∗ , Ruihan Wu ∗, Zhiyuan Zhang, Kamalika Chaudhuri
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-04697

Browse the full library · About the Tower of Babel