Privacy-Aware Time-Series Data Sharing With Deep Reinforcement Learning

Internet of things (IoT) devices are becoming increasingly popular thanks to many new services and applications they offer. However, in addition their benefits, raise privacy concerns since share fine-grained time-series user data with untrusted third parties. In this work, we study the privacy-utility trade-off (PUT) sharing. Existing approaches PUT mainly focus on a single point; however, temporal correlations introduce challenges. Methods that preserve for current time may leak significant amount information at trace level as adversary can exploit trace. We consider sharing distorted version user's true sequence an party. measure leakage by mutual between shared version. both instantaneous average distortion two sequences, under given measure, utility loss metric. To tackle history-dependent minimization, reformulate problem Markov decision process (MDP), solve it using asynchronous actor-critic deep reinforcement learning (RL). evaluate performance proposed solution location synthetic GeoLife GPS trajectory datasets. For latter, show validity our testing released against network.