Practical and Robust Federated Learning With Highly Scalable Regression Training.

Area of Science:

• Computer Science
• Machine Learning
• Cybersecurity

Background:

• Federated learning (FL) enables machine learning model training without centralizing raw data, crucial for privacy in Internet of Medical Things (IoMT).
• Traditional interactive FL schemes face privacy and security threats due to multiple communication rounds.
• Existing noninteractive FL schemes struggle with local data privacy, scalability, dropout tolerance, and result verification.

Purpose of the Study:

• To propose novel noninteractive federated learning (NFRT) schemes for IoMT that address existing challenges in privacy, scalability, robustness, and verifiability.
• To develop practical NFRT solutions balancing privacy preservation with efficient and secure regression model training.

Main Methods:

• Introduced two NFRT schemes: Homomorphic Encryption based NFRT (HE-NFRT) and Double-Masking Protocol based NFRT (Mask-NFRT).
• Both schemes are designed for privacy-preserving computation, high-efficiency, robustness against data owner dropout, and include a verification mechanism.
• Security and performance analyses were conducted to evaluate the proposed methods.

Main Results:

• The proposed HE-NFRT and Mask-NFRT schemes effectively protect the privacy of local datasets for data owners (DOs).
• Both schemes demonstrate resilience against collusion attacks and provide strong verification capabilities for each DO.
• HE-NFRT is suitable for high-dimensional, high-security IoMT applications, while Mask-NFRT is optimized for high-dimensional, large-scale IoMT scenarios.

Conclusions:

• The developed NFRT schemes offer practical solutions for privacy-preserving federated learning in IoMT.
• These methods enhance security, scalability, and verifiability in distributed machine learning for healthcare applications.
• The choice between HE-NFRT and Mask-NFRT depends on specific IoMT application requirements regarding security and scale.