Stackelberg Game Based Performance Optimization in Digital Twin Assisted Federated Learning over NOMA Networks

Despite its advantage of preserving data privacy, federated learning (FL) could suffer from the limited computation resources of the distributed clients particularly when they are connected by wireless networks. By imitating the distributed resources effectively, digital twin (DT) shows great potential in eliminating the straggler issue in FL. In this paper, we leverage DT in the FL framework over non-orthogonal multiple access (NOMA) network, where DT deployed at the server can assist FL training process. To minimize the total latency and energy consumption in the proposed system, we formulate a Stackelberg game by considering clients and the server as the leader and the follower, respectively. Specifically, the leader aims to minimize the energy consumption via the optimization of DT mapping data ratio and resource allocation, while the objective of the follower is to minimize the total latency during FL training by optimally allocating DT computation resource. The Stackelberg equilibrium is considered to obtain the optimal solutions. We first derive the closed-form solution for the follower-level problem and include it in the leader-level problem which is then solved through the deep reinforcement learning (DRL) method. Simulation results verify the superior performance of the proposed scheme.