Privacy-preserving Analysis of Correlated Data

Many modern databases include personal and sensitive cor-related data, such as private information on users connectedtogether in a social network, and measurements of physicalactivity of single subjects across time. However, differentialprivacy, the current gold standard in data privacy, does notadequately address privacy issues in this kind of data.This work looks at a recent generalization of differentialprivacy, called Pufferfish, that can be used to address privacyin correlated data. The main challenge in applying Pufferfishis a lack of suitable mechanisms. We provide the first mech-anism – the Wasserstein Mechanism – which applies to anygeneral Pufferfish framework. Since this mechanism may becomputationally inefficient, we provide an additional mech-anism that applies to some practical cases such as physicalactivity measurements across time, and is computationallyefficient. Our experimental evaluations indicate that thismechanism provides privacy and utility for synthetic as wellas real data in two separate domains.