Locally differentially private estimation of functionals of discrete distributions

Abstract

We study the problem of estimating non-linear functionals of discrete distributions in the context of local differential privacy. The initial data x_1,,x_n [K] are supposed i.i.d. and distributed according to an unknown discrete distribution p = (p_1,,p_K). Only -locally differentially private (LDP) samples z_1,...,z_n are publicly available, where the term 'local' means that each z_i is produced using one individual attribute x_i. We exhibit privacy mechanisms (PM) that are interactive, i.e. they are allowed to use already published confidential data, or non-interactive. We describe the behavior of the squared error risk in estimating the power sum F_ = _k=1^K p_k^, >0 as a function of K, \, n and . In the non-interactive case, we study several plug-in type estimators of F_, for all >0, that are similar to the MLE which has been analyzed by Jiao et al. (2017) in the multinomial model. However, due to the privacy constraint the rates we attain are slower and similar to those obtained in the Gaussian model by Collier et al. (2020). In the interactive case, we introduce for all >1 a two-step procedure which attains the faster parametric rates (n ^2)^-1/2 when 2. We give lower bounds results over all -LDP privacy mechanisms and over all estimators using the private samples.

Tasks

Reproductions