Energy Efficient Federated Learning with Age-Weighted FedSGD

This paper investigates federated learning in a wireless communication system, where random device selection is employed with non-independent and identically distributed (non-IID) data distribution. In order to mitigate the weight divergence issue, age-weighted federated stochastic gradient descent (FedSGD) is designed, where a weighting factor is introduced to scale local gradients according to the previous state of the device. Furthermore, under the maximum time constraint, an energy consumption minimization problem is formulated to increase the participation of devices. By transforming the proposed problem into convex and utilizing KKT conditions, the optimal resource allocation solution is derived. Simulation results indicate that age-weighted FedSGD is able to outperform conventional FedSGD in terms of convergence rate and achievable accuracy, and the proposed resource allocation strategy can further improve the learning performance by increasing device availability.