Dynamic Digital Twin and Federated Learning with Incentives for Air-Ground Networks

Abstract

The air-ground network provides users with seamless connections and real-time services, while its resource constraint triggers a paradigm shift from machine learning to federated learning. Federated learning enables clients to collaboratively train models without sharing data. Meanwhile, digital twins provide environmental awareness and autonomous management, which in combination with federated learning reconciles the conflict between privacy protection and data training in air-ground network. In this paper, we consider dynamic digital twin and federated learning for air-ground networks where drone works as the aggregator and the ground clients collaboratively train the model based on the network dynamics captured by digital twins. We design incentives for federated learning based on Stackelberg game, in which the digital twin of the drone acts as the leader to set preferences for clients, and clients as followers choose the global training rounds after weighing benefits and costs. Furthermore, considering the varying digital twin deviations and network dynamics during the federated learning process, we design a dynamic incentive scheme to adaptively adjust the selection of the optimal clients and their participation level. Numerical results show that the proposed schemes can significantly improve accuracy and energy efficiency.