Continuous Location Statistics Sharing Algorithm with Local Differential Privacy

Abstract

Continuous sharing of location statistics produces valuable knowledge to understand important phenomena, such as popular places or pattern behaviors. Most importantly, data should be shared without jeopardizing users' privacy. Differential privacy becomes de-facto technique for private statistical data release. Much work focuses on the centralized setting where users send their original data to a trusted server. Then the server adds controlled noises to generate differentially private statistics. This centralized approach is vulnerable to attacks where an adversary may access the true data by attacking the trusted server. Local differential privacy neutralizes this type of attacks by allowing each user to obfuscate their data before it reaches the server for statistical analysis. In this paper, we propose an algorithm to share location statistics that leverages local differential privacy combined with w-event privacy. Our solution guarantees the user's privacy when continuously releasing statistics over infinite streams. Experimental evaluation on real-life data shows our solution with strong privacy guarantee.