VFLA: A Verifiable Federated Learning Scheme for Multi-Round Aggregation in VANETs

In Vehicular Ad-hoc Networks (VANETs), federated learning (FL) enables collaborative training of a global intelligent traffic model without sharing raw data. Model convergence is achieved through multiple rounds of FL. The latest work, Flamingo (S&P 2023), proposes a multi-round aggregation scheme that effectively improves training efficiency. However, in this scheme, vehicles are unable to verify the aggregation results, and computational errors by the semi-trusted server can reduce model accuracy and lead to training failure. This scheme proposes a verifiable FL scheme for multi-round aggregation in VANETs (VFLA). Firstly, a multi-round verification mechanism is designed that reduces vehicle interactions by using reusable keys. Additionally, we propose a lightweight proof scheme that allows vehicles to verify results with minimal computation, thereby reducing the overhead of the verification process. Finally, we analyze the security in the case of vehicle dropout and experimentally validate the efficiency of VFLA on different datasets. Compared to Flamingo, the time overhead shows no significant increase, and the verification time is reduced by approximately 90% compared to the state-of-the-art scheme, demonstrating the scheme’s usability.