Edge-Intelligence-Based Computation Offloading Technology for Distributed Internet of Unmanned Aerial Vehicles
Wenhua Wang, Yilin Zhang, Qin Liu, Tian Wang, Weijia Jia
Abstract
With the development of networks and smart devices, artificial intelligence has drawn more and more attention, especially in the Unmanned Aerial Vehicles. Therefore, it is quite critical to train and run DNNs on resource-limited and hardware-constrained UAVs. The traditional methods fail to adjust offloading strategy due to the dynamic environment, while recently proposed intelligent computation offloading techniques rely on accessing IoT devices’ private data, which leads to privacy and security problem. To alleviate the above problems, we propose an novel edge-intelligent-based computation offloading technology via Federated Learning (FL). Specially, we utilize Multi-Layer Perceptron (MLP) to learn the computation tasks features and offload different tasks to different smart devices. Besides, to protect data privacy and improve the system’s security, a hierarchical FL framework is utilized to train the model of the computation tasks features extraction. Finally, performance analysis results obtained by experiments demonstrate the performance of our proposed approach.