Improving the Number of Queries Supported by Differentially Private Mechanisms
Wen Cai Huang, Shijie Zhou, Yongjian Liao · 2021
Differential privacy is a widely accepted privacy conception. However, there is a limitation on the number of queries supported by differentially private mechanisms, and the limitation stops differentially private mechanisms from being widely used in business systems. In this paper, we propose two mechanisms to optimize the number of queries, namely approximate answer differentially private mechanism and nonzero mean differentially private mechanism. We analyze the two proposed mechanisms in terms of their privacy and utility. On the one hand, the two mechanisms satisfy E differential privacy. On the other hand, outputs of proposed mechanisms are indistinguishable over an interval so the number of queries is improved greatly. For utility, we show that the utility of the two proposed mechanisms is the same as that of the Laplace mechanism in the view of information theory.