Energy-Efficient Federated Learning Over UAV-Enabled Wireless Powered Communications

Abstract

Since the invention in 2016, federated learning (FL) has been a key concept of artificial intelligence, in which the data of FL users needs not to be uploaded to the central server. However, performing FL tasks may not be feasible due to the unavailability of terrestrial communications and the battery limitation of FL users. To address these issues, we make use of unmanned aerial vehicles (UAVs) and wireless powered communications (WPC) for FL networks. In order to enable sustainable FL solutions, the UAV equipped with edge computing and WPC capabilities is deployed as an aerial energy source as well as an aerial server to perform FL tasks. We propose a joint algorithm of UAV placement, power control, transmission time, model accuracy, bandwidth allocation, and computing resources, namely energy-efficient FL (E2FL), aiming at minimizing the total energy consumption of the aerial server and users. The E2FL overcomes the original nonconvex problem by an efficient algorithm. We show that sustainable FL solutions can be provided via UAV-enabled WPC through various simulation results. Moreover, the outperformance of E2FL in terms of energy efficiency over several benchmarks emphasizes the need for a joint resource allocation framework rather than optimizing a subset of optimization factors.