Differentially private set intersection cardinalities in the multiparty setting

Abstract

This paper examines the problem of learning information on multiple private databases in a quick and secure manner. The owners of such databases may either not desire to or are otherwise unable to leak information about their respective databases, despite the desired information possibly being entirely benign in nature. For example, to reference an earlier paper creating the DJoin system, one may wish to study the correlation between disease and travel to certain regions of the world, but not have access to the information required for this process. This project builds upon past attempts at providing a system to answer this problem, with a focus on improving the scalability as the number of datasets being examined increases.

As such, this paper introduces the Secure Multiparty Differentially Private Mechanism, or MDP, designed to improve upon past works in this field, and offer a universal solution to the problem that works regardless of the number of parties involved.