CS700:Graduate Seminar in Computer Science & Informatics

Real-Time Aggregate Monitoring with Differential Privacy
Liyue Fan, Department of Mathematics and Computer Science

Sharing real-time aggregate statistics of private data has given much benefit to the public to perform data mining for understanding important phenomena, such as Influenza outbreaks and traffic patterns. Our study is aimed to enable data holders, i.e. hospitals and wireless service providers, to continuously share accurate real-time aggregates under differential privacy, a rigorous privacy guarantee. In this talk, I will discuss the challenges in sharing single time-series as well as spatial/multidimensional time series. As the proposed solution, I will present FAST, a real-time system with Filtering and Adaptive Sampling for sharing Time series with differential privacy. FAST adaptively samples long time-series according to detected data dynamics and simultaneously uses filtering to dynamically predict and correct data values. For multidimensional time series, FAST uses spatial decomposition techniques combined with sampling and filtering to improve utility. I will briefly go over each component in FAST and will show empirical studies with both synthetic data and real-world data sets which confirm that FAST improves the accuracy of released data series compared to existing methods.