Sync-GWO: Highly Private and Bandwidth-Efficient Federated Learning With a Case Study in Healthcare.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Distributed Systems

Background:

• Federated Learning (FL) enables collaborative model training while preserving data privacy.
• Traditional FL methods like FedAvg face challenges with high communication overhead and privacy risks from gradient sharing.

Purpose of the Study:

• To present a novel FL framework using Multi-Objective Optimization (MOO).
• To introduce Sync-GWO, an adaptation of the Grey Wolf Optimizer (GWO) for efficient and private FL.

Main Methods:

• Formulating federated optimization as a MOO problem.
• Developing Sync-GWO, which uses synchronized Pseudo-Random Number Generators (PRNGs) to avoid population data transmission.
• Testing the framework on COVID-19 pandemic-related Internet of Medical Things (IoMT) data.

Main Results:

• Sync-GWO reduces communication costs by over 99% (to ~1KB/round).
• Achieves up to 15% higher accuracy and 2.5x improved F1-scores compared to FedAvg.
• Demonstrates robust performance on imbalanced datasets and non-differentiable objectives.

Conclusions:

• Sync-GWO offers significant improvements in communication efficiency and privacy for FL.
• The method is particularly suitable for privacy-sensitive and communication-constrained applications like IoMT.
• Sync-GWO provides a viable alternative to gradient-based FL methods, especially in challenging optimization scenarios.