Multi-owner multi-user privacy

In the realm of Internet of Things, sensitive information is distributed among several data owners, while multiple data users wish to access different aspects of this information. This paper presents an approach for a multi-owner multi-user (MOMU) system where data owners require privacy guarantees before offering their private data. In such a setting each owner has different privacy needs against each user, whereas, users may seek to collaborate in order to violate owners' privacy. Using approximate differential privacy, we focus on the case where n data owners possess a real-valued private data and m data users wish to learn a linear query of this data. We consider a Gaussian mechanism, derive the constraints on the covariance matrix for the mechanism to be multi-owner multi-user private, and propose a convex semi-definite relaxation to design the covariance. Finally, we illustrate our approach to a synthetic scenario where n agents act both as data owners and data users and we evaluate the privacy and the accuracy of the resulted mechanism.