Local Differential Privacy for Physical Sensor Data
Physical sensors (thermal, light, motion, etc.) are becoming ubiquitous and offer important benefits to society. However, allowing sensors into our private spaces has resulted in considerable privacy concerns. Differential privacy has been developed to help alleviate these privacy concerns. In this talk, we'll develop and define a framework for releasing physical data that preserves both utility and provides privacy. Our notion of closeness of physical data will be defined via the Earth Mover Distance and we'll discuss the implications of this choice. Physical data, such as temperature distributions, are often only accessible to us via a linear transformation of the data. We'll analyze the implications of our privacy definition for linear inverse problems, focusing on those that are traditionally considered to be "ill-conditioned". We'll then instantiate our framework with the heat kernel on graphs and discuss how the privacy parameter relates to the connectivity of the graph. Our work indicates that it is possible to produce locally private sensor measurements that both keep the exact locations of the heat sources private and permit recovery of the "general geographic vicinity" of the sources. Joint work with Anna C. Gilbert.
Audra McMillan is a final year PhD candidate in the Department of Mathematics at the University of Michigan under the advisement of Anna C. Gilbert. She is interested in machine learning and inverse problems, with a focus on privacy-preserving data analysis. She received her B.Sc in mathematics from the University of Sydney.