Differentially private high dimensional sparse covariance matrix estimation

DOI10.1016/J.TCS.2021.03.001MaRDI QIDQ2661783zbMATH OpenOpenAlexFDO

Authors Di Wang, Jinhui Xu

Publication date 8 April 2021

Published in Theoretical Computer Science (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1901.06413

zbMATH Keywords

differential privacyhigh-dimensional statisticssparse covariance estimation

Mathematics Subject Classification ID

Estimation in multivariate analysis (62H12) Privacy of data (68P27)

Abstract: In this paper, we study the problem of estimating the covariance matrix under differential privacy, where the underlying covariance matrix is assumed to be sparse and of high dimensions. We propose a new method, called DP-Thresholding, to achieve a non-trivial $el{l}_2$-norm based error bound, which is significantly better than the existing ones from adding noise directly to the empirical covariance matrix. We also extend the $el{l}_2$-norm based error bound to a general $el{l}_w$-norm based one for any $1leqwleqinfty$, and show that they share the same upper bound asymptotically. Our approach can be easily extended to local differential privacy. Experiments on the synthetic datasets show consistent results with our theoretical claims.

Recommendations

• Tight lower bound of sparse covariance matrix estimation in the local differential privacy model
• Differentially private precision matrix estimation
• The cost of privacy: optimal rates of convergence for parameter estimation with differential privacy
• Principal component analysis in the local differential privacy model

Cites work

• An introduction to matrix concentration inequalities
• Analyze Gauss: optimal bounds for privacy-preserving principal component analysis
• Covariance regularization by thresholding
• High-dimensional probability. An introduction with applications in data science
• scientific article; zbMATH DE number 3137662 (Why is no real title available?)
• scientific article; zbMATH DE number 3165002 (Why is no real title available?)
• scientific article; zbMATH DE number 6026126 (Why is no real title available?)
• Optimal rates of convergence for covariance matrix estimation
• Optimal rates of convergence for sparse covariance matrix estimation
• Our Data, Ourselves: Privacy Via Distributed Noise Generation
• Principal component analysis in the local differential privacy model
• Theory of Cryptography
• What can we learn privately?

Cited in

(7)

• Principal component analysis in the local differential privacy model
• Tight lower bound of sparse covariance matrix estimation in the local differential privacy model
• Differentially private precision matrix estimation
• Smooth sensitivity based approach for differentially private principal component analysis
• On differentially private low rank approximation
• Covariance loss, Szemeredi regularity, and differential privacy
• Distributed empirical likelihood inference with privacy guarantees

Describes a project that uses

Uses Software

• RAPPOR

This page was built for publication: Differentially private high dimensional sparse covariance matrix estimation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2661783)