Joseph Near, University of California, Berkeley

Abstract

Differential privacy promises to enable general data analytics while protecting individual privacy. However, applications of differential privacy in industry (e.g. at Google and Apple) have so far focused on collecting specific and limited kinds of information from users.

In this talk, I will discuss the development of a differentially private interface for general-purpose database queries. In collaboration with Uber, we developed a set of requirements for deploying such an interface in a practical setting--among them, support for joins and the ability to work with an existing database. I will describe our approach, elastic sensitivity, for satisfying these requirements, and detail our experience implementing the approach in Uber's setting.

Slides     Video