With the widespread deployment of multi-sensors on vehicles, a significant amount of data is generated that exhibits the characteristics of massive volume, wide variety, and privacy sensitivity. The private information protection poses immense challenges to intelligent applications in intelligent transportation systems (ITS). Federated learning (FL) has been introduced to support advanced ITS applications with minimum data exchanges and privacy disclosure. However, besides achieving data privacy in such a system, the malicious behavior or unintentional misbehavior of participating vehicles caused by high vehicle mobility and limited resources are critical considerations that may hinder the adoption of FL in ITS. In this paper, the concept of reputation is introduced as a metric to evaluate the reliability and trustworthiness of vehicles' behavior. Additionally, the reputation evaluation mechanism of the joint vehicle mobility metric and local model update performance metric is designed to calculate the reputation score of the vehicle at each model aggregation round. Furthermore, considering the case that well-behaved vehicles are misjudged due to mobility, we propose a spot-check strategy to verify and employ low-reputation but reliable misjudged vehicles with a certain probability based on their reputation value to improve the model training efficiency. Extensive experiments are conducted on real traffic signal datasets to demonstrate the effectiveness of our proposed scheme.
Reliable Federated Learning with Mobility-Aware Reputation Mechanism for Internet of Vehicles
2023-09-24
786703 byte
Conference paper
Electronic Resource
English
FedBeam: Reliable Incentive Mechanisms for Federated Learning in UAV-Enabled Internet of Vehicles
DOAJ | 2024
|