Air-Ground Integrated Federated Learning: An Experimental Implementation

With the envision of sixth generation (6G) networks technology, diverse artificial intelligence (AI) services are gradually developed from the network center to the edge, which enables unmanned aerial vehicle (UAV) as a hot spot to provide auxiliary services of machine learning (ML). Due to the sensitive privacy and limited resources, traditional centralized ML may not be used directly in such networks. As a promising distributed collaborative ML, federated learning (FL) is more suitable. However, the current research on air-ground integrated federal learning (AGIFL) mostly focuses on architecture design and network simulation. In this paper, we mainly verify the feasibility of deploying FL between UAV and terrestrial users through practical platform. Extensive experiments demonstrate that the training accuracy is higher when the training nodes is close to the server, and the accuracy is improved from 69.48% to 83.52% when distance is reduced from 20 meters to 5 meters.