Private Hypothesis Testing via Robustness ---- Audra McMillan

Hypothesis testing plays a central role in statistical inference, and is used in many settings where privacy concerns are paramount. In this talk, we will discuss the role that robustness plays in designing good differentially private hypothesis tests. We’ll focus on two foundational problems: simple hypothesis testing and identity testing in high dimensions. Both problems are challenging to solve privately, for slightly different reasons. We’ll see how improving the robustness of non-private tests allows us to design private algorithms that adapt to the specific instance they are given. We’ll also discuss the role that dimensionality plays in private hypothesis testing in high dimensions.