Minimaxity, Statistical Thinking and Differential Privacy

It is important that privacy methodology be supported by rigorous theory. The purpose of this paper is to introduce researchers in privacy to some statistical theory that is relevant for privacy. The emphasis is on the role of statistical minimax theory and differential privacy. I will discuss some differences between how statisticians think about these issues versus how computer scientists think about them. To large extent, this paper is more of an essay, offering some general ideas and open problems, and suggesting avenues for future research. Similar viewpoints were taken in previous work, for example, Dwork and Smith (2010) and Wasserman and Zhou (2010).