Enhancing Trust by a Keycloak-Flower Integration for Federated Machine Learning

Area of Science:

• Computer Science
• Medical Informatics
• Cybersecurity

Background:

• Federated learning (FL) shows promise in healthcare but has limited real-world application.
• Existing FL frameworks like Flower need enhanced security for sensitive medical data.
• Identity and access management are critical for secure FL deployment in healthcare organizations.

Purpose of the Study:

• To improve the security and access management of federated learning in healthcare.
• To integrate Keycloak, a robust identity and access management solution, with the Flower FL framework.
• To address security gaps in FL applications within healthcare organizations.

Main Methods:

• Developed a lightweight Python module to integrate Keycloak with gRPC and Flower.
• Implemented code validation on the server-side before client code execution.
• Tested the integrated system in a basic prototype environment.

Main Results:

• Successful integration of Keycloak with gRPC and Flower for enhanced identity and access management.
• Demonstrated a mechanism for server-side validation of client code within the FL framework.
• A functional prototype was established, validating the proposed integration approach.

Conclusions:

• The proposed integration enhances FL security in healthcare by improving identity and access management.
• Further research and comprehensive security testing are necessary for complex, real-world healthcare deployments.
• This approach offers a pathway to more secure and widespread adoption of federated learning in the medical field.