Federated Learning AI
Master
Semester programme:Art-IE
Huyen Thai
Meda Benissa
Hristo Kolev
Mario Petkov
Boris Marinov
Dimitar Barev
Project description
Building up a software system for managing and monitoring communication between client and central server.
Performing a research on multi-tenancy.
Context
Federated Learning (FL) is a machine learning method that allows different parties to train AI models together without sharing their raw data, keeping information private and reducing the need for central data storage. The Art-IE Federated Learning Lab, part of the Art-IE project at Fontys ICT Center of Expertise AI, aims to promote FL by encouraging collaboration between universities and businesses, especially small and medium-sized companies in Flanders and the Netherlands. This project focuses on solving key challenges like communication between systems, differences in data, and security, helping to create stronger AI models while protecting data privacy.
Results
This study explored how to design a federated learning platform that meets the goals of enterprise scalability and regulatory compliance. Four key design pillars were identified:
- Modular architecture: Using microservices on AKS enables horizontal scalability and service isolation.
- Performance and resource management: Auto-scaling, telemetry and graceful degradation allow the system to maintain performance under heavy loads.
- Privacy protection: Federated learning is inherently privacy-first, but this is enhanced with SMPC, differential privacy and federated quotas.
- Compliance integration: Mapping the platform’s features to GDPR, HIPAA and ISO standards ensures legal robustness and audit readiness.
By embedding scalability, observability and compliance into the foundation of the system, the proposed platform addresses the core challenges of enterprise FL deployments. While some components need validation, the architecture is designed to grow and scale with both technical and regulatory requirements in mind.
The result is a future ready platform that helps organizations train AI models collaboratively without compromising performance, privacy or compliance.