Mobile Devices Strategies in Blockchain-Based Federated Learning: A Dynamic Game Perspective
Sizheng Fan, Hongbo Zhang, Zehua Wang, Wei Cai
Abstract
Leveraging various mobile devices to train the shared model collaboratively, federated learning (FL) can improve the privacy and security of 6G communication. To economically encourage the participation of heterogeneous mobile devices, an incentive mechanism and a fair trading platform are needed. In this paper, we implement a blockchain-based FL system and propose an incentive mechanism to establish a decentralized and transparent trading platform. Moreover, to better understand the mobile devices’ behaviors, we provide economic analysis for this market. Specifically, we propose two strategy models for mobile devices, namely the discrete strategy model (DSM) and the continuous strategy model (CSM). Also, we formulate the interactions among the non-cooperative mobile devices as a dynamic game, where they adjust their strategies iteratively to maximize the individual payoff based on others’ previous strategies. We further prove the existence in Nash equilibrium (NE) of two different models and propose algorithms to achieve them. Simulation results demonstrate the convergence of the proposed algorithms and show that the CSM can effectively increase the mobile devices’ payoffs to 128.1 percent at most compared with DSM.