RaftFed: An Efficient Federated Learning Framework for Vehicular Crowd Intelligence

Vehicular crowd intelligence (VCI) is an emerging research field. Facilitated by state-of-the-art vehicular ad-hoc networks and artificial intelligence, various VCI applications come to place, e.g., collaborative sensing, positioning, and mapping. The collaborative property of VCI applications generally requires data to be shared among participants, thus forming network-wide intelligence. How to fulfill this process without compromising data privacy remains a challenging issue. Although federated learning (FL) is a promising tool to solve the problem, adapting conventional FL frameworks to VCI is nontrivial. First, the centralized model aggregation is unreliable in VCI because of the existence of stragglers with unfavorable channel conditions. Second, existing FL schemes are vulnerable to NonIID data, which is intensified by the data heterogeneity in VCI. Third, the dynamic nature of vehicular networks necessitates robust handling of nodes frequently opting in and out. This paper proposes a novel dual-layer federated learning framework called RaftFed to facilitate efficiently privacy-preserving VCI. By integrating with heartbeat and cache mechanisms, RaftFed is endowed with dynamic and asynchronous capabilities. The experimental results also show that RaftFed performs better than baselines regarding communication overhead, model accuracy, and model convergence.