Princeton University

School of Engineering & Applied Science

Information Theoretic Metrics for Security and Privacy

Speaker: 
Flavio du Pin Calmon
Location: 
EQuad B205
Date/Time: 
Wednesday, December 3, 2014 - 4:30pm

There is a complementarity between inference and security (privacy). In this talk, we explore this relationship by combining tools from information theory, statistics, and estimation theory in order to develop new metrics for quantifying information leakage in security systems.  We use these new tools to (i) characterize the optimal privacy-utility tradeoff in privacy-preserving mechanisms, and (ii) determine the fundamental performance limits of symmetric-key encryption. Furthermore, we introduce security metrics and associated results based on the spectrum of the conditional expectation operator, called the principal inertia components. The principal inertia components allow a fine-grained decomposition of the dependence between a hidden and an observed variable which, in turn, is useful for deriving fundamental bounds for estimation problems, and for measuring information leakage in secure communication models. Finally, we illustrate how our results can be used as a design driver for practical security and privacy schemes. 
 
Bio: Flavio du Pin Calmon is a PhD candidate at the Electrical Engineering and Computer Science Department at MIT, and a member of the Network Coding and Reliable Communications Group at the Research Laboratory of Electronics (RLE). His research interests include information theory, statistics, security and privacy. In addition to his work at MIT, Flavio has ongoing collaborations with the MIT Lincoln Laboratory and Technicolor SA. Before coming to MIT, he received an M.Sc. in Electrical Engineering from the Universidade Estadual de Campinas, Brazil, and a B.Sc. in Communications Engineering from the Universidade de Brasilia, Brazil.