FedGraphHE: A privacy-preserving federated graph neural network framework with dynamic homomorphic encryption and...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Network Security

Background:

• Federated learning (FL) allows collaborative AI model training across devices without sharing raw data.
• Graph neural networks (GNNs) are effective for analyzing complex relationships in healthcare data.
• Existing federated GNNs struggle with privacy vulnerabilities, high computational costs, and Byzantine attacks.

Purpose of the Study:

• To develop FedGraphHE, a privacy-preserving federated GNN framework for secure collaborative intelligence in smart healthcare.
• To address gradient privacy, computational overhead, and Byzantine attack challenges in federated GNNs.

Main Methods:

• Integration of three synergistic modules: Dynamic Adaptive Partitioned Homomorphic Encryption (DAPHE) for optimized gradient transmission.
• Hierarchical Multi-scale Adaptive Graph Transformer (HMAGT) for encryption-aware graph processing.
• Federated Robust Aggregation via Homomorphic Inner Product (FRAHIP) for Byzantine-resilient aggregation.

Main Results:

• FedGraphHE outperforms existing privacy-preserving methods on citation network benchmarks (Cora, CiteSeer, PubMed).
• Achieved 98.18% classification accuracy on medical imaging datasets (ISIC 2020).
• Reduced communication costs by ~25% compared to homomorphic encryption baselines and maintained >95% accuracy under Byzantine attacks.

Conclusions:

• FedGraphHE offers an effective solution for privacy-sensitive collaborative learning in healthcare.
• The framework enhances diagnostic accuracy and security in smart healthcare networks.
• Demonstrates significant improvements in performance, efficiency, and robustness over existing methods.