Multi-Objective Distributed Client Selection in Federated Learning-Assisted Internet of Vehicles

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Federated learning (FL) is a distributed machine learning framework increasingly used in the Internet of Vehicles (IoV).
• Millions of vehicles participate in FL to train models, but maintaining active states incurs significant costs due to the large number of participants.
• The need for efficient client selection in IoV federated learning is critical to manage computational and communication overhead.

Purpose of the Study:

• To propose a distributed client selection scheme for federated learning in the Internet of Vehicles.
• To reduce the cost associated with maintaining active states for a large number of participating vehicles.
• To develop an effective client evaluation mechanism that balances multiple critical variables.

Main Methods:

• A distributed client selection scheme is proposed, where clients are elected among neighbors based on their evaluation.
• An evaluator considers four key variables: sample quantity, available throughput, computational capability, and local dataset quality.
• Fuzzy logic is adopted as the evaluation method due to the lack of a closed-form solution for the multi-variable problem.

Main Results:

• The proposed distributed client selection scheme effectively reduces the cost of maintaining active states for participants.
• The method approximates the accuracy of centralized client selection methods.
• Extensive simulations demonstrate a significant reduction in communication overhead.

Conclusions:

• The developed distributed client selection scheme is efficient for federated learning in the Internet of Vehicles.
• Fuzzy logic provides a robust approach for evaluating clients based on multiple, complex variables.
• The findings suggest a practical solution for optimizing federated learning in large-scale vehicular networks.