Divergence-Private Query Release and High-Dimensional Mean Estimation

Overview

This paper studies how power-divergence privacy governs the release of statistical queries and the noise needed to protect neighboring datasets. It connects privacy mechanisms with the statistical problem of estimating a high-dimensional mean.

Methods and contribution

• Develops sensitivity-based Gaussian noise calibration for query release under power-divergence privacy.
• Compares Gaussian and Laplace mechanisms under Hellinger privacy.
• Analyzes high-dimensional mean estimation, including clipping for unbounded data.

Publication

Fengnan Deng, Anand N. Vidyashankar. Divergence-Private Query Release and High-Dimensional Mean Estimation. Statistics & Probability Letters, 2026. Volume 239, article 110918.