Department Seminar by Kelly Ramsay

Please Note: This seminar will be given online.

Introduction to Differential Privacy

The goal of the talk is to give a gentle introduction to differential privacy and present some of my recent research related to private inference. Differential privacy is a popular standard of privacy recently employed by many large institutions, notably the US census bureau and Google. We touch briefly on the origins of differential privacy before discussing the definition and some basic building blocks for constructing private algorithms. We emphasize that several connections between differential privacy and robust statistics have been identified in the literature. Our recent work explores one such connection: privatizing data depth functions. We present some general methods for privatizing depth functions and their associated medians. We also analyse the utility-privacy trade-off with some asymptotic results and present a general extension of a recent version of the propose-test-release paradigm.