On Purely Private Covariance Estimation
Abstract
We present a simple perturbation mechanism for the release of -dimensional covariance matrices under pure differential privacy. For large datasets with at least elements, our mechanism recovers the provably optimal Frobenius norm error guarantees of \cite{nikolov2023private}, while simultaneously achieving best known error for all other -Schatten norms, with . Our error is information-theoretically optimal for all , in particular, our mechanism is the first purely private covariance estimator that achieves optimal error in spectral norm. For small datasets , we further show that by projecting the output onto the nuclear norm ball of appropriate radius, our algorithm achieves the optimal Frobenius norm error , improving over the known bounds of of \cite{nikolov2023private} and of \cite{dong2022differentially}.
Keywords
Cite
@article{arxiv.2510.26717,
title = {On Purely Private Covariance Estimation},
author = {Tommaso d'Orsi and Gleb Novikov},
journal= {arXiv preprint arXiv:2510.26717},
year = {2026}
}
Comments
ALT 2026; equal contribution