Differentially private data publishing for arbitrarily partitioned data

Abstract

Many models have been proposed to preserve data privacy for different data publishing scenarios. Among these models, ∊-differential privacy is receiving increasing attention because it does not make assumptions about adversaries’ prior knowledge and can provide a rigorous privacy guarantee. Although there are numerous proposed approaches using ∊-differential privacy to publish centralized data of a single-party, differentially private data publishing for distributed data among multiple parties has not been studied extensively. The challenge in releasing distributed data is how to protect privacy and integrity during collaborative data integration and anonymization. In this paper, we present the first differentially private solution to anonymize data from two parties with arbitrarily partitioned data in a semi-honest model. We aim at satisfying two privacy requirements: (1) the collaborative anonymization should satisfy differential privacy; (2) one party cannot learn extra information about the other party’s data except for the final result and the information that can be inferred from the result. To meet these privacy requirements, we propose a distributed differentially private anonymization algorithm and guarantee that each step of the algorithm satisfies the definition of secure two-party computation. In addition to the security and cost analyses, we demonstrate the utility of our algorithm in classification analysis.