Differentially Private Model-X Knockoffs via Johnson-Lindenstrauss Transform

Yuxuan Tao Adel Javanmard · 2025

Read Differentially Private Model-X Knockoffs via Johnson-Lindenstrauss Transform by Yuxuan Tao Adel Javanmard (2025) 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-2508-04800.txt.

Description

Yuxuan Tao and Adel Javanmard present a private version of model-X knockoffs for controlled variable selection, in a research paper from 2025. Knockoffs build fake negative-control features to keep false discovery rates in check. Standard privacy tools, such as added Gaussian noise, break the swap property that makes knockoffs valid, so a new route is needed.

Their route privatizes the data matrix with a Johnson-Lindenstrauss transform, a projection that roughly preserves distances. Theory shows how privacy level, sample size, and feature count shape the trade-off between privacy and power. A debiasing step aids the analysis. Conditions are given for power to reach full strength over time.

Catalog record

CREATORYuxuan Tao Adel Javanmard
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE192 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-04800

Browse the full library · About the Tower of Babel