Deim Seminar
Title
Privacy in Distributed Networks: Information Theory vs. Estimation Theory
Conferenciant
Shabab Asoodeh
Professor/a organitzador/a
Josep Domingo-Ferrer
Institution
Queen's University, Canada
Date
18-07-2016 12:00
Summary
Motivated by applications in social networks, we talk about different measures of privacy and utility using information-theoretic and estimation-theoretic quantities. Informally, we define privacy and utility as the amount of information that is leaked into the "displayed data" about the "private data" and "non-private data", respectively. We then formulate the underlying conflict between utility and privacy by introducing the "privacy-utility function" and also "estimation noise-to-signal ratio". Despite their intuitive definitions, these two functions have the following shortcomings; (1) they are very difficult to be computed, unless the joint distribution of private data and non-private data satisfies a notion of symmetry, and (2) they lack information-theoretic "operational" interpretations. Then to (partially) overcome these problems, we propose a new privacy-utility tradeoff which is both intuitive and information-theoretically operational when both private and non-private data have discrete distributions. We then talk about the properties of this new quantity and show how this generalizes the previous results in this area.
Place
Laboratori 231
Language
Anglès
