UAV-Assisted Federated Learning With Robust Resource and Trajectory Optimization Under Location Uncertainties

Federated learning (FL) has emerged as a promising solution to facilitate the deployment of artificial intelligence (AI) on wireless devices. However, heterogeneity of wireless devices, including disparities in computation capabilities, data sizes, and energy constraints, introduces delays in the FL completion time, particularly due to inefficient communication and slow updates from resource-constrained devices. To address this issue, we propose an unmanned aerial vehicle (UAV)-assisted FL framework that integrates UAV as a central server, collaborating with the devices to facilitate the model training process. Accordingly, we jointly consider computation and transmission strategies, as well as the task assignment and UAV trajectory to minimize the completion time of the FL process. Particularly, we consider the location uncertainties associated with the devices, along with the consequent chance-constrained aggregation process, to achieve a robust learning process. We employ the Bernstein-type inequalities to reformulate the probabilistic-form optimization into its deterministic counterpart. Then we solve the problem under a block coordinate descent framework. Simulation results demonstrate that the proposed approach significantly reduces the completion time of FL and achieves robust performance guarantee in the presence of location deviations.