Federated Learning with Channel and Energy Aware Scheduling

In this paper, a federated learning setup in which multiple devices capable of harvesting energy from the environment train a machine learning model based on the intermittent availability of the energy and channel is studied. The main focus is on developing an algorithm that achieves the same convergence as state-of-the-art federated learning methods in a scenario with an error-prone channel and intermittent energy availability. We propose a federated learning algorithm that schedules distributed clients and weighting their local gradients according to the energy and channel profiles of each client. The performance of the proposed algorithm has been demonstrated with the experiments.