Federated Learning with Pareto Optimality for Resource Efficiency and Fast Model Convergence in Mobile Environments.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Federated learning (FL) enables model training on decentralized data, preserving user privacy.
• Existing FL methods face challenges including network latency, limited user device resources, and parameter server congestion.
• These limitations hinder efficient resource utilization and rapid model convergence.

Purpose of the Study:

• To propose a novel resource-efficient federated learning scheme.
• To address limitations of traditional FL regarding network and computing resource consumption.
• To enhance model convergence speed and reduce network load.

Main Methods:

• Implemented a resource-efficient FL scheme incorporating Pareto optimality and biased client selection.
• Utilized a hierarchical structure with location-based clustering for device-to-device (D2D) communication.
• Employed k-means clustering for efficient device grouping.

Main Results:

• The proposed scheme significantly reduced transmitted and received network traffic by 75.89% and 78.77%, respectively, compared to FedAvg at a 0.75 participation rate.
• Achieved faster model convergence compared to FedAvg and D2D-FedAvg.
• Demonstrated efficient resource consumption and management.

Conclusions:

• The proposed resource-efficient FL scheme effectively mitigates network congestion and resource limitations.
• Hierarchical clustering and biased client selection enhance FL performance and efficiency.
• This approach offers a promising solution for large-scale, resource-constrained federated learning applications.