Federated edge learning with reconfigurable intelligent surface and its application in Internet of vehicles

Abstract

Aiming at the problem that it is difficult to achieve an optimal trade-off between wireless communication and model accuracy in FEEL training caused by the heterogeneity of wireless links and data distribution, a reconfigurable intelligent surface (RIS) enabled federated edge learning system was proposed, which exploited the channel reconfigurability of RIS to adaptively manipulate the signal propagation environment, and utilized the over-the-air computation (Aircomp) to achieve fast model aggregation.Specifically, the convergence behavior of the FEEL algorithm under the influence of wireless channels and data heterogeneity was rigorously derived, and accordingly, an unified wireless resources optimization problem was constructed with the goal of minimizing the learning loss by jointly designing the transceiver design and the RIS phase shift.Simulation results demonstrate that the proposed scheme achieves substantial performance improvement compared against several baselines, and prove that RIS can play an important role in improving the accuracy of Aircomp enabled FEEL systems under data heterogeneity.Finally, the probability of applying it into Internet of vehicles is discussed.

Keywords

• federated edge learning;data heterogeneity;reconfigurable intelligent surface;over-the-air computation
• convergence analysis;Internet of vehicles