SIoT Selection, Clustering, and Routing for Federated Learning with Privacy-Preservation

Min-Siou Chung, Chih-Hang Wang, De-Nian Yang, Guang-Siang Lee, Wen-Tsuen Chen, Jang-Ping Sheu

Abstract: With the advances in Social Internet of Things (SIoT) and Federated learning (FL), smart devices are now able to cooperatively and locally perform learning tasks to protect sensitive data by Differential Privacy (DP). On the other hand, Hierarchical FL (HFL) clusters SIoTs into multiple local training groups to reduce communication overheads by local aggregation. In this paper, we explore SIoT Training Group Construction (STGC) for HFL to minimize the total SIoT computation, communication and hiring costs, and the privacy cost for exploiting DP. We prove that STGC is NP-hard and inapproximable within any factor unless P = NP. Then, we design an algorithm with the ideas of Coverage Efficiency Indicator, Data Balance-aware Dual Adjustment, and Privacy-Aware Rerouting to choose and cluster SIoTs and to determine the aggregator for local training and SIoT routing in each cluster. Simulation results manifest that the proposed algorithm outperforms state-of-the-arts regarding the total cost, model accuracy, and convergence time.