Differentially private PCA algorithm works without eigenvalue gaps

Gap-free Differentially Private PCA for Gaussian Data

Data Structures and AlgorithmsMachine Learning

Summary

When trying to find the main directions of variation in data (called principal components) while keeping individual data private, existing methods often need clear differences between these directions. This paper presents a way to do this with Gaussian data even when these differences are small or absent. The authors provide a new algorithm that maintains privacy without relying on gaps in the data. This could help analyze sensitive data more accurately and privately.

What this means in practice

  • For data privacy engineers: Design systems that extract main data features while ensuring privacy guarantees even when data patterns lack clear separations.
  • For medical data analysts: Analyze sensitive Gaussian-distributed medical data to identify key factors without requiring large separations in data characteristics and while protecting patient confidentiality.

Tested on simulated data.

Authors

Alina Ene, Huy L. Nguyen

Abstract

We give a gap-free differentially private algorithm for the principal component analysis (PCA) problem with Gaussian data.