Architecting a differentially private SQL engine
In recent years, differential privacy (DP) has emerged as the state-of-the-art for privately analyzing sensitive data. Despite its wide acceptance in the academic community and much work on differentially private algorithm design, there is surprisingly little work on building database systems that allow differentially private query answering using high level, declarative languages like SQL. The lack of such systems has limited the adoption of differential privacy in real-world applications. In this paper, we propose PRIVSQL, a system architecture for supporting SQL query answering under differential privacy and identify a set of components that can be independently optimized. While there is a mature class of solutions for some components, there is little or no work for others. Our preliminary implementation can support a richer class of SQL queries than a state of the art competitor, with accuracy that is as much as 7000× better.
