Differentially-Private Fingerprinting of Relational Databases
share
When sharing sensitive databases with other parties, a database owner aims to (i) have privacy guarantees for its shared database, (ii) have liability guarantees in case of unauthorized sharing of its database by the recipients, and (iii) provide a high quality (utility) database to the recipients. We observe that sharing a database under differential privacy and database fingerprinting are orthogonal objectives. The former aims to inject noise into a database to prevent inference of the original data values, whereas, the latter aims to hide unique marks inside a database to trace malicious parties who leak the data without the authorization. In this paper, we achieve these two objectives simultaneously by proposing a novel differentially-private fingerprinting mechanism for databases. Specifically, we first devise a bit-level random response scheme to achieve differential privacy for sharing entire databases, and then, based on this, we develop an ϵ-differentially private fingerprinting mechanism. Next, we theoretically analyze the relationships among differential privacy guarantee, fingerprint robustness, and database utility by deriving closed form expressions to characterize the privacy-utility coupling and privacy-fingerprint robustness coupling. Furthermore, we propose a sparse vector technique (SVT)-based solution to control the cumulative privacy loss when fingerprinted copies of a database are shared with multiple recipients. We experimentally show that our mechanism achieves stronger fingerprint robustness than the state-of-the-art fingerprinting mechanisms, and higher database utility than the simple composition of database perturbation under differential privacy followed by fingerprinting (e.g., statistical utility of the shared database by the proposed scheme is more than 10x higher than perturbation followed by fingerprinting).
READ FULL TEXT