Cluster Head Selection and Cluster Construction for Federated Learning in Mobile Ad Hoc Networks

Federated learning (FL), as a distributed learning framework, which can effectively protect user privacy and address data imbalance issues. In this paper, we investigate cluster head (CH) selection and client clustering issues based on hierarchical FL framework in mobile ad hoc networks (MANETs), wherein all the clients are randomly distributed without infrastructure support. Specifically, we formulate the joint client association and CH selection problem, with the target of minimizing the training latency while maximizing the system lifetime. Furthermore, we propose a heuristic algorithm for client association and an entropy weight Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method for CH selection. Simulation results demonstrate that our proposed algorithms can effectively shorten training latency and extend lifetime compared with other schemes.