A data sharing method for internet of drones based on federated learning

Abstract

The transmission of massive data between connected drones makes it an important issue to ensure the efficient data sharing and user privacy security. The introduction of federated learning can effectively solve the problem of privacy protection. However, it is difficult to maintain a continuous and stable synchronous communication mechanism in the process of training. There are still problems of data redundancy and low sharing efficiency. Thus, this paper proposes a Multi-level Asynchronous Federated Learning (MAFL) architecture, realizing efficient data sharing in the Internet of Drones (IoD). The MAFL architecture deploys different federated learning training participation strategies for different IoD entities. The drone entity is deployed with the distributed local training by using initiative inquiry mechanism. For the edge entity of IoD, on the one hand, it enhances the performance of drone local training by designing data set delivery scheme; On the other hand, the weighted aggregation training is deployed to improve the convergence speed and accuracy of MAFL. For the central entity of IoD, the global aggregation training is deployed to accelerate the synchronization of MAFL. The simulation analysis shows that the MAFL can support training of different entities with fast convergence and high accuracy in IoD.

Keywords: Asynchronous training, Data sharing, Federated learning, IoD