Differentially Private Multi-dimensional Time Series Release for Traffic Monitoring

Sharing real-time traffic data can be of great value to understanding many important phenomena, such as congestion patterns or popular places. To this end, private user data must be aggregated and shared continuously over time with data privacy guarantee. However, releasing time series data with standard differential privacy mechanism can lead to high perturbation error due to the correlation between time stamps. In addition, data sparsity in the spatial domain imposes another challenge to user privacy as well as utility. To address the challenges, we propose a real-time framework that guarantees differential privacy for individual users and releases accurate data for research purposes. We present two estimation algorithms designed to utilize domain knowledge in order to mitigate the effect of perturbation error. Evaluations with simulated traffic data show our solutions outperform existing methods in both utility and computation efficiency, enabling real-time data sharing with strong privacy guarantee.