Mobile computing power trading decision-making method for vehicle-mounted devices in multi-task edge federated learning

Abstract

With the development of edge computing and artificial intelligence technology, edge federated learning (EFL) has been widely applied in the Internet of Vehicles (IOV) due to its distributed characteristics and advantages in privacy protection. In this paper, we study the mobile computing power trading between edge servers (ES) and mobile vehicle-mounted equipment (MVE) in the IOV scene. In order to reduce the influence of MVEs’ flexibility, which can easily lead to single point failure or offline problem, we propose semi-synchronous FL aggregation. Considering that multiple federated learning (FL) tasks have different budgets and MVEs have different computing resources, we design an incentive mechanism to encourage selfish MVEs to actively participate in FL task training, so as to obtain higher quality FL models. Furthermore, we propose a fast association decision method based on dynamic state space Markov decision process (DSS-MDP). Simulation experiment data show that, MVEs can obtain higher quality local models at the same energy consumption, thus gaining higher utility. Semi-synchronous FL aggregation is able to improve the accuracy of FL global model by 0.764% on average and reduce the idle time of MVEs by 90.44% compared with the way of allocating aggregation weights according to the data volume.