CIS Theory Seminar Series: Jinshuo Dong

PCPSE (Perelman Center), room 200

January 24, 2020

12:00 PM - 1:00 PM

Title: How private are private algorithms?

Abstract: It is important to understand the exact privacy guarantee provided by private algorithms developed in the past prosperous decade of differential privacy. In particular, any underestimation of actual privacy means unnecessary noise in the algorithm and loss in the final accuracy. We observe a central limit behavior in iterative private algorithms, which demonstrates the limit of the common $(\varepsilon,\delta)$ parametrization in most application scenarios. For the rescue, a new notion called Gaussian Differential Privacy (GDP) is proposed and a complete toolkit is developed. We carry out various experiments to show how much unnecessary loss of accuracy can be saved in deep learning applications. Based on joint work with Aaron Roth, Weijie Su, Zhiqi Bu and Qi Long.

Bio: Jinshuo Dong is a PhD student in the Applied Mathematics and Computational Science (AMCS) program at the University of Pennsylvania, advised by Aaron Roth. He obtained his bachelor’s degree in mathematics from Peking University in 2014. His main research is on differential privacy and its role in statistics and machine learning.