Energy-Efficient Federated Learning-Based Device Selection in IoT Environments

This article presents the Energy-Time Efficient Device Selection (ETEDS) algorithm for net federation IoT devices, for an AI-reliant environment. The novel approach focuses on device selection in each Federated Learning (FL) round, considering IoT devices' computation capacity and proximity to the BS, aiming to limit energy use, preserve communication time, and maintain high accuracy of the FL method. It formulates the device selection problem as an optimization task and provides a heuristic algorithm to solve it. The ETEDS algorithm identifies convergence points and suggests stopping training at specific epochs, leading to significant energy and time savings while maintaining satisfactory accuracy levels. Experimental results demonstrate the superiority of the ETEDS algorithm in terms of energy efficiency and overall performance in dynamic IoT scenarios, showcasing approximately 10% savings in total energy and time consumption.