Concentrating on the Foundations of Differential Privacy — Thomas Steinke

Abstract: Differential privacy has, over the past decade, developed a rich theory and a burgeoning practice. This talk re-examines the fundamental definitions of differential privacy that underlie its success. In particular, I will discuss concentrated differential privacy and its variants which use Rényi divergence to quantify privacy loss, instead of directly bounding probabilities. This approach provides sharper and more natural composition bounds, while mostly retaining the other key properties of differential privacy. I will argue that this provides a useful and practical analytical tool, while also providing a valuable theoretical perspective on differential privacy.

Bio: Thomas Steinke is a postdoctoral researcher at the IBM Almaden Research Center working on privacy and learning theory. He recently completed his PhD at Harvard university advised by Salil Vadhan. Prior to that he was an undergraduate student in New Zealand.