English

On Purely Private Covariance Estimation

Machine Learning 2026-02-03 v2 Data Structures and Algorithms

Abstract

We present a simple perturbation mechanism for the release of dd-dimensional covariance matrices Σ\Sigma under pure differential privacy. For large datasets with at least nd2/εn\geq d^2/\varepsilon elements, our mechanism recovers the provably optimal Frobenius norm error guarantees of \cite{nikolov2023private}, while simultaneously achieving best known error for all other pp-Schatten norms, with p[1,]p\in [1,\infty]. Our error is information-theoretically optimal for all p2p\ge 2, in particular, our mechanism is the first purely private covariance estimator that achieves optimal error in spectral norm. For small datasets n<d2/εn< d^2/\varepsilon, we further show that by projecting the output onto the nuclear norm ball of appropriate radius, our algorithm achieves the optimal Frobenius norm error O(d  Tr(Σ)/n)O(\sqrt{d\;\text{Tr}(\Sigma) /n}), improving over the known bounds of O(d/n)O(\sqrt{d/n}) of \cite{nikolov2023private} and O(d3/4Tr(Σ)/n){O}\big(d^{3/4}\sqrt{\text{Tr}(\Sigma)/n}\big) 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

R2 v1 2026-07-01T07:14:14.370Z