A Link Adaptive Approach for Federated Learning aided UAV Networks

Federated learning (FL) assisted unmanned aerial vehicle (UAV) networks have been considered as a promising technology for the future Internet of Things (IoT) scenarios, where both the communication cost and computational cost can be reduced. In this paper, a FL assisted UAV networks are considered for performing image classification tasks in IoT, where an aerial base station (ABS) is deployed as the aggregation center and UAVs are deployed as the local entities. Firstly, the system performance is analyzed in terms of algorithm convergence and energy consumption. Then, we propose a link adaptive approach for the FL assisted UAV networks, where a link quality selection scheme and a model parameter selection scheme are conceived at the ABS and each UAV, respectively. Finally, the system performance is evaluated by simulation results, showing that the proposed link adaptive approach is capable of attaining a faster convergence performance and a higher classification accuracy compared to the conventional FL scheme.