Paper
Privacy Loss of Noise Perturbation via Concentration Analysis of A Product Measure
Authors
Shuainan Liu, Tianxi Ji, Zhongshuo Fang, Lu Wei, Pan Li
Abstract
Noise perturbation is one of the most fundamental approaches for achieving -differential privacy (DP) guarantees when releasing the result of a query or function evaluated on a sensitive dataset . In this approach, calibrated noise is used to obscure the difference vector , where is known as a neighboring dataset. A DP guarantee is obtained by studying the tail probability bound of a privacy loss random variable (PLRV), defined as the Radon-Nikodym derivative between two distributions. When follows a multivariate Gaussian distribution, the PLRV is characterized as a specific univariate Gaussian. In this paper, we propose a novel scheme to generate by leveraging the fact that the perturbation noise is typically spherically symmetric (i.e., the distribution is rotationally invariant around the origin). The new noise generation scheme allows us to investigate the privacy loss from a geometric perspective and express the resulting PLRV using a product measure, ; measure is related to a radius random variable controlling the magnitude of , while measure involves a directional random variable governing the angle between and the difference . We derive a closed-form moment bound on the product measure to prove -DP. Under the same -DP guarantee, our mechanism yields a smaller expected noise magnitude than the classic Gaussian noise in high dimensions, thereby significantly improving the utility of the noisy result . To validate this, we consider convex and non-convex empirical risk minimization (ERM) problems in high dimensional space and apply the proposed product noise to achieve privacy.