Mean Estimation with User-level Privacy under Data Heterogeneity
Abstract
A key challenge in many modern data analysis tasks is that user data are heterogeneous. Different users may possess vastly different numbers of data points. More importantly, it cannot be assumed that all users sample from the same underlying distribution. This is true, for example in language data, where different speech styles result in data heterogeneity. In this work we propose a simple model of heterogeneous user data that allows user data to differ in both distribution and quantity of data, and provide a method for estimating the population-level mean while preserving user-level differential privacy. We demonstrate asymptotic optimality of our estimator and also prove general lower bounds on the error achievable in the setting we introduce.
Cite
@article{arxiv.2307.15835,
title = {Mean Estimation with User-level Privacy under Data Heterogeneity},
author = {Rachel Cummings and Vitaly Feldman and Audra McMillan and Kunal Talwar},
journal= {arXiv preprint arXiv:2307.15835},
year = {2023}
}
Comments
Conference version published at NeurIPS 2022