Communication-Efficient Private Distribution Estimation via Combinatorial Designs
Abstract
In today's world, our diverse information is collected through various channels and utilized for a range of purposes, including statistical inference and the development of machine learning models. However, privacy threats continue to emerge, revealing that sensitive personal information can be inferred from statistics or machine learning models. To address this, it is necessary to appropriately randomize data during the collection stage to protect privacy, but this inevitably reduces data utility. This talk will explore the optimal privacy-utility trade-off in the context of distribution estimation, a representative statistical inference task. Specifically, we will examine how classical results in combinatorial design can be effectively leveraged to achieve this trade-off in a communication-efficient manner.
Biography
Si-Hyeon Lee is an Associate Professor in the School of Electrical Engineering at KAIST, South Korea. Before joining KAIST EE, she was an Assistant Professor in the Department of Electrical Engineering, POSTECH, South Korea. She received the B.S. (summa cum laude) and Ph.D. degrees in Electrical Engineering from KAIST, in 2007 and 2013, respectively. From 2014 to 2016, she was a Postdoctoral Fellow in the Department of Electrical and Computer Engineering at the University of Toronto, Canada. Her research interests include information theory, wireless communications, statistical inference, and machine learning.