Federated Learning-Based Task Offloading in a UAV-Aided Cloud Computing Mobile Network

Abstract

Unmanned aerial vehicles (UAVs), due to their flexibility in deployment, offer various advantages in the next-generation wireless networks. In this work, we study a UAV-assisted mobile edge computing network where a UAV is equipped with computing resources. The user equipment (UE) is capable of transferring a part of the computational workload to the UAV. We aim to reduce the maximum processing delay while adhering to the energy consumption limitation. This is accomplished by optimizing user scheduling, task offloading ratio, UAV's flight angle, and flight speed. We propose a computation offloading approach using federated learning based on decentralized federated averaging, taking into account the fact that this optimization problem is not convex, the state space is high-dimensional, and the action space is continuous. We determine the best strategy for offloading computing in an environment whose dynamics are challenging to regulate. We perform extensive simulation testing, and our findings indicate that the proposed method converges faster to the optimal solution. Furthermore, the proposed algorithm significantly improves the processing delay by about 20% at smaller task sizes and even higher for larger task sizes as compared to the baseline algorithms.