Mobility-Aware Federated Learning Considering Multiple Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Federated learning (FL) enables distributed machine learning (ML) model training while preserving user data ownership.
• User mobility and network disconnections can significantly disrupt FL training efficiency, leading to client dropouts.

Purpose of the Study:

• To propose MoFeL, a novel FL coordination algorithm designed to maintain efficient training performance in the presence of user mobility.
• To evaluate MoFeL's effectiveness in handling multiple networks and varying central server configurations.

Main Methods:

• Developed and simulated the MoFeL coordination algorithm for federated learning.
• Utilized an image classification application with a convolutional neural network for experimental evaluation.
• Compared MoFeL's performance against traditional FL coordination algorithms in high-mobility scenarios.

Main Results:

• MoFeL demonstrated superior performance in FL training coordination under high mobility conditions.
• The proposed algorithm achieved 156.5% more training cycles compared to a non-mobility-aware algorithm.
• MoFeL effectively managed training across multiple networks with different central servers.

Conclusions:

• MoFeL significantly enhances the efficiency and robustness of federated learning in dynamic, mobile environments.
• The algorithm offers a practical solution for overcoming client dropout challenges in real-world FL deployments.
• MoFeL represents a substantial advancement in distributed machine learning coordination strategies.