A Many-objective Joint Device Selection and Aggregation Scheme for Federated Learning in IoV

Advanced mobile communication and data processing technologies have promoted the development of Internet of Things (IoT), but they have also posed challenges to the distributed federated learning mode in the field of Internet of Vehicles (IoV). Faced with a large number of vehicle nodes available for federated training in IoV, the federated learning training task becomes challenging when motivating a large number of vehicles participant. A difficulty posed for federated learning in IoV is heterogeneity challenges caused by massive device participation. Moreover, excessive resource and system maintenance costs associated with a large number of poor-quality devices participating in federated training cannot be ignored. To address these issues, this paper proposes a novel vehicle device selection and aggregation joint optimization model based on a many-objective evolutionary algorithm. The proposed model can be optimized by BiGE algorithm to obtain an optimal subset of vehicle equipment and corresponding weight assignment scheme, thus reducing unnecessary resources waste and budget expenditure while ensuring global model performance. To verify the feasibility of the model, several sets of experiments are conducted to demonstrate that our proposed model has acceptable performance while largely reducing the number budget of participants.