Differentially private data aggregation with optimal utility

Computing aggregate statistics about user data is of vital importance for a variety of services and systems, but this practice has been shown to seriously undermine the privacy of users. Differential privacy has proved to be an effective tool to sanitize queries over a database, and various cryptographic protocols have been recently proposed to enforce differential privacy in a distributed setting, e.g., statical queries on sensitive data stored on the user's side. The widespread deployment of differential privacy techniques in real-life settings is, however, undermined by several limitations that existing constructions suffer from: they support only a limited class of queries, they pose a trade-off between privacy and utility of the query result, they are affected by the answer pollution problem, or they are inefficient. This paper presents PrivaDA, a novel design architecture for distributed differential privacy that leverages recent advances in secure multiparty computations on fixed and floating point arithmetics to overcome the previously mentioned limitations. In particular, PrivaDA supports a variety of perturbation mechanisms (e.g., the Laplace, discrete Laplace, and exponential mechanisms) and it constitutes the first generic technique to generate noise in a fully distributed manner while maintaining the optimal utility. Furthermore, PrivaDA does not suffer from the answer pollution problem. We demonstrate the efficiency of PrivaDA with a performance evaluation, and its expressiveness and flexibility by illustrating several application scenarios such as privacy-preserving web analytics.