Implementing federated learning over VPN-based wireless backhaul networks for healthcare systems

Area of Science:

• Computer Science
• Telecommunications Engineering
• Cybersecurity

Background:

• Federated learning (FL) enables collaborative machine learning model training on edge devices.
• Analyzing sensitive healthcare records requires efficient and secure network solutions.
• Wireless backhaul networks present unique challenges for FL implementation.

Purpose of the Study:

• To introduce an efficient federated learning network for healthcare record analysis.
• To evaluate the performance of different wireless backhaul channels for FL.
• To enhance the security of federated learning networks in healthcare.

Main Methods:

• Implemented a federated learning approach using Virtual Private Network (VPN) technology over a wireless backhaul.
• Compared Terahertz (THz), E/V band (mmWave), and microwave communication channels.
• Assessed model accuracy and convergence time using the FedAvg algorithm.
• Developed a three-step multi-layered security strategy.

Main Results:

• The proposed FL network with VPN over wireless backhaul demonstrated superior performance.
• Terahertz (THz) communication yielded the highest accuracy and fastest convergence (38 seconds vs. 55 seconds).
• The multi-layered security approach integrated private networks, licensed frequency channels, and VPNs.

Conclusions:

• Faster communication links significantly improve federated learning network efficiency.
• The combination of THz communication, VPNs, private networks, and licensed channels offers a robust and secure solution for healthcare data analysis.
• This study highlights the potential of optimized wireless backhaul for secure and efficient federated learning in critical sectors.