Federated deep reinforcement learning based secure data sharing for Internet of Things

摘要

The increasing number of Internet of Things (IoT) devices motivate the data sharing that improves the quality of IoT services. However, data providers usually suffer from the privacy leakage caused by direct data sharing. To solve this problem, in this paper, we propose a Federated Learning based Secure data Sharing mechanism for IoT, named FL2S. Specifically, to accomplish efficient and secure data sharing, a hierarchical asynchronous federated learning (FL) framework is developed based on the sensitive task decomposition. In addition, to improve data sharing quality, the deep reinforcement learning (DRL) technology is utilized to select participants of sufficient computational capabilities and high quality datasets. By integrating task decomposition and participant selection, reliable data sharing is realized by sharing local data models instead of the source data with data privacy preserved. Experiment results show that the proposed FL2S achieves high accuracy in secure data sharing for various IoT applications.