SensorSift: balancing sensor data privacy and utility in automated face understanding

Abstract

We introduce SensorSift, a new theoretical scheme for balancing utility and privacy in smart sensor applications. At the heart of our contribution is an algorithm which transforms raw sensor data into a 'sifted' representation which minimizes exposure of user defined private attributes while maximally exposing application-requested public attributes. We envision multiple applications using the same platform, and requesting access to public attributes explicitly not known at the time of the platform creation. Support for future-defined public attributes, while still preserving the defined privacy of the private attributes, is a central challenge that we tackle. To evaluate our approach, we apply SensorSift to the PubFig dataset of celebrity face images, and study how well we can simultaneously hide and reveal various policy combinations of face attributes using machine classifiers. We find that as long as the public and private attributes are not significantly correlated, it is possible to generate a sifting transformation which reduces private attribute inferences to random guessing while maximally retaining classifier accuracy of public attributes relative to raw data (average PubLoss = .053 and PrivLoss = .075, see Figure 4). In addition, our sifting transformations led to consistent classification performance when evaluated using a set of five modern machine learning methods (linear SVM, kNearest Neighbors, Random Forests, kernel SVM, and Neural Nets).