Unmanned aerial vehicles (UAVs) need to deploy machine learning (ML) models to execute complex tasks such as target tracking. For traditional centralized ML, UAVs need to transmit the raw data to the ground central server for model training. However, the transmission of raw data not only causes a large delay but also may incur sensitive information leakage. Federated learning (FL) as a distributed machine learning enables multiple UAVs to form a swarm for distributed model training. Nevertheless, most of current researches on FL of UAV swarm are based on theoretical architecture and simulations. In this paper, we focus on verifying the feasibility of deploying FL in a UAV swarm through practical experiments. Moreover, we further explore the impact of the distance change between the training node UAVs and the parameter server UAV on the training accuracy. The experimental results show that the UAV swarm can perform 15 rounds of global training in 5 minutes, with a training accuracy of 84.28%.
Experimental Results of Federated Learning in Unmanned Aerial Vehicle Swarm
2021-10-01
519449 byte
Conference paper
Electronic Resource
English