Edge Intelligence: Federated Learning-Based Privacy Protection Framework for Smart Healthcare Systems

Area of Science:

• Artificial Intelligence
• Cybersecurity
• Healthcare Technology

Background:

• Federated learning (FL) offers privacy in Internet of Things (IoT) networks, especially smart healthcare.
• Centralized FL aggregation is vulnerable to privacy attacks and manipulation, compromising sensitive patient data and model efficiency.
• Existing methods struggle to balance robust privacy protection with high model performance in edge computing environments.

Purpose of the Study:

• To introduce Edge Intelligence, a novel federated learning-based privacy protection framework for edge-based smart healthcare systems.
• To enhance data security and safeguard against privacy attacks like inference and free-riding in sensitive healthcare data.
• To develop a framework that balances stringent privacy preservation with high model accuracy and efficiency.

Main Methods:

• Implementation of a three-fold Federated Edge Aggregator framework named Edge Intelligence.
• Utilization of an iteration-based Conventional Neural Network (CNN) model integrated with artificial noise functions.
• Theoretical analysis of Edge Intelligence's convergence bound on the federated learning model's loss function.

Main Results:

• The proposed Edge Intelligence framework demonstrated strong performance across diverse datasets (MNIST, CIFAR10, STL10, COVID-19 chest x-ray).
• Achieved an average accuracy of 90% while maintaining a high privacy budget.
• Outperformed established baseline methods in both model performance and privacy protection capabilities.

Conclusions:

• Edge Intelligence offers a robust and effective solution for privacy protection in smart healthcare systems at the edge.
• The framework successfully balances the critical trade-off between privacy preservation and model performance.
• This approach represents a significant advancement in securing sensitive health data within federated learning paradigms.