Double Auction Mechanism for Cooperative Swarm Learning in Internet of Vehicles

Abstract

Federated Learning (FL) plays a prominent role for realizing distributed machine learning in the Internet of Vehicles (IoV). However, the clients still need to transfer their local models to the central coordinator in FL, which brings redundant communication overhead. In Swarm Learning (SL), there exists no central coordinator and all joint members can merge others’ parameters through a swarm network. This paper proposes a cooperative SL framework that improves training efficiency and security. To incentivize vehicle users (VUs) to cooperatively train, we further design an iterative double auction based SL scheme to maximize the total social welfare and stimulate VUs with good channel states and sufficient resources to participate in SL. The extensive experiment results show the effectiveness of the proposed scheme.