Underestimating the power of inference attacks is the major reason why data privacy mechanisms fail. In this talk, I will describe my general approach to quantifying privacy and illustrate its applications by showing how to rigorously measure privacy risks of location data and machine-learning models. I will then discuss my current research at the junction of privacy and data science in an important practical scenario: generating privacy-preserving synthetic data while preserving utility of real data.

Reza Shokri is a postdoctoral researcher at Cornell Tech. His research focuses on quantitative analysis of privacy, as well as design and implementation of privacy technologies for practical applications. His work on quantifying location privacy was recognized as a runner-up for the Award for Outstanding Research in Privacy Enhancing Technologies (PET Award). Recently, he has focused on privacy-preserving generative models and privacy in machine learning. He received his PhD from EPFL.