Privately Answering Counting Queries with Generalized Gaussian Mechanisms
Abstract
We consider the problem of answering counting (i.e. sensitivity-1) queries about a database with -differential privacy. We give a mechanism such that if the true answers to the queries are the vector , the mechanism outputs answers with the -error guarantee: This reduces the multiplicative gap between the best known upper and lower bounds on -error from to . Our main technical contribution is an analysis of the family of mechanisms of the following form for answering counting queries: Sample from a \textit{Generalized Gaussian}, i.e. with probability proportional to , and output . This family of mechanisms offers a tradeoff between and -error guarantees and may be of independent interest. For , this mechanism already matches the previous best known -error bound. We arrive at our main result by composing this mechanism for with the sparse vector mechanism, generalizing a technique of Steinke and Ullman.
Cite
@article{arxiv.2010.01457,
title = {Privately Answering Counting Queries with Generalized Gaussian Mechanisms},
author = {Arun Ganesh and Jiazheng Zhao},
journal= {arXiv preprint arXiv:2010.01457},
year = {2020}
}