Princeton University

School of Engineering & Applied Science

Balancing behavioral privacy and information utility while sharing sensory data streams

Supriyo Chakraborty, IBM T. J. Watson Research Center
Friend Center 108
Thursday, December 3, 2015 - 4:30pm to 5:30pm

Abstract:  Our personal, social, work, and urban spaces are being increasingly instrumented using sensors of various modalities, resulting in the generation and sharing of large amounts of personal sensor data. Embedded in this data are digital footprints of our daily lives, and, therefore, the problem of protecting the behavioral privacy of users while simultaneously maintaining the utility of the shared data has become more relevant than ever before.
Traditional approaches to addressing privacy threats have primarily focused on data sanitization for disassociating the user's identity from shared data and on anonymization techniques to make a user indistinguishable within a sub-population. However, due to the nature of applications in domains such as mobile health, insurance, etc., user identity is often an inalienable part of the shared data. Thus, instead of identity privacy, we focus on protecting the privacy of inferences that can be derived from the shared data.
In this talk, we will present a formalization of the notion of inference privacy for sensor data, discuss privacy metrics and algorithms for achieving inference privacy, and outline a system architecture for their realization.