Federated Learning Empowered Resource Allocation in UAV-Assisted Edge Intelligent Systems

Abstract

Mobile edge computing (MEC) has been considered a promising advanced technology to support delay-sensitive tasks of user equipment (UE) in the internet of things (IoT) systems, it is necessary to allow multiple UEs to offload their computationally intensive tasks to a flexible edge computing server, such as an unmanned aerial vehicle (UAV)-assisted edge computing server. However, most existing works mainly focused on minimising energy consumption under the transmission and/or processing delay constraints while ignoring privacy-preserving, which will be challenging when dealing with large volumes of raw data. In this paper, we consider a federated learning (FL) empowered UAV-assisted edge intelligent system to minimise the maximum utility cost (which indicates the relationship between latency and energy consumption) to the selected UE for task processing. We propose to jointly optimise the FL task offloading decisions among all UEs and the communication resource allocation under each epoch. This is achieved by devising a federated learning-based edge intelligence offloading decision optimisation algorithm (FEOA). Simulation results show that our proposed schemes outperform the benchmarks in terms of the maximum cost efficiency among all UEs.