Privacy-Aware Task Offloading via Two-Timescale Reinforcement Learning

Mobile-edge computing (MEC) enables low-latency computing services by by deploying the computing resources at the logical edge of the network and allowing mobile users to wirelessly offload their computation-intensive tasks. Meanwhile, as user privacy is receiving increasing attention in the modern society, mitigating the privacy leakage caused by task offloading in MEC becomes imperative. In this paper, we develop a reinforcement learning (RL) based privacy-aware task offloading scheme that can synthetically take into account the data privacy, the usage pattern privacy, and the location privacy of the mobile users. To find the optimal offloading strategy, a novel two-timescale RL algorithm, dubbed as statistic prediction-post decision state-virtual experience (SP-PDS-VE), is proposed. The proposed algorithm can construct the state transition model of the underlying problem via the fast timescale learning and, in the meantime, uses the learned model to create a set of virtual experience for the slow timescale learning, so as to speed up the convergence and allow the mobile device to learn the optimal privacy-aware offloading strategy much faster. In addition to the analysis, simulations results are presented to corroborate the effectiveness of the proposed scheme.