Abstract

Data privacy has been put onfirm theoretical foundations since the birth of differential privacy. Thefruitful decade has witnessed many theoretical and practical successes,including the remarkable deployment in Chrome and iOS. However, a majorobstacle to further applications is the extra sophistication in varioustheorems in differential privacy, most notably composition and subsampling.Precise characterization of these properties is of vital importance inpractice. We propose a hypothesis testing based framework called f-DP, whichincludes Gaussian differential privacy (GDP) as the most useful special case.All desired properties admit elegant and tight statements in this framework. Inaddition, a central limit theorem is proved and shows the universality of GDP.As a comprehensive application, we show how the tools we develop improves theanalysis of the privacy of perturbed stochastic gradient descent.

Based on joint work with AaronRoth and Weijie Su.

Time

2月26日（周二）15:00-16:30

Speaker

Jinshuo Dong is aPhD student in the applied math program at University of Pennsylvania,supervised by Aaron Roth. Before coming to Penn, he received bachelor's degreein Mathematics from Peking University in 2014.

Venue

信息管理与工程学院102室

上海财经大学（第三教学楼西侧）

上海市杨浦区武东路100号