Abstract: Differential privacy (DP) has been one of the most widely studied and used frameworks for characterising privacy leakage from data-sharing systems in both academia and industry. However, the leakage characteristics of differential privacy, especially in terms of Bayesian inference attacks, are not fully understood. In this talk, we utilise concepts from information theory to give an operational interpretation of the epsilon parameter in the local differential privacy (LDP) framework.
Our characterisation establishes that epsilon in LDP is indeed an attainable upper bound on the worst-case Bayesian inference leakage from any data-sharing system.
Bio: Parastoo Sadeghi received bachelor’s and master’s degrees in electrical engineering from the Sharif University of Technology, Tehran, Iran, in 1995 and 1997, respectively, and a Ph.D. degree in electrical engineering from the University of New South Wales (UNSW), Sydney, in 2006. Parastoo is a Senior Member of IEEE and is currently a Professor at the School of Engineering and Information Technology, UNSW Canberra.
She has co-authored over 200 refereed journal articles and conference papers. Her research interests include information theory, data privacy, index coding, and network coding. From 2016 to 2019, she served as an Associate Editor for the IEEE Transactions on Information Theory. In 2022, she was selected as a Distinguished IEEE Information Theory Society Lecturer. She is currently an ARC Future Fellow and is also serving as the Secretary of the IEEE Information Theory Society.
Disclaimer: This video was recorded when the speaker gave a seminar at Digital Futures, a cross-disciplinary research centre that explores and develops digital technologies of great societal importance (www.digitalfut...). The content is produced by the speaker and not the center. The captions have been generated automatically. If you need refined captions to access the material, please reach out to the Associate Director for Seminars at Digital Futures.
Негізгі бет Parastoo Sadeghi - Viewing Differential Privacy through the Lens of Information Theory
Пікірлер