Secure and Accurate Federated Learning as a Service (SA-FLaaS)

We enable privacy preserving federated learning for medical imaging AI

Duration of the project

1.8.2024 – 31.7.2026

Source of funding

Business Finland

Total funding

636 723 €

SA FLaaS is a platform that enables training high performing machine learning models without moving sensitive data. It addresses key federated learning challenges by combining security, accuracy, privacy preservation, and regulatory compliance in a single solution. SA FLaaS allows organizations to develop advanced machine learning models while retaining full control over their data.

The project builds on the PRIVASA project (2021-23), which presented award-winning FL solutions for federated tumor segmentation in Federated Learning (FL) is a decentralized machine learning paradigm that enables collaborative model training across distributed data providers while keeping data local and preserving privacy. Despite its advantages in data protection and security, FL faces significant challenge. SA FLaaS addresses these challenges by providing a secure and scalable federated learning platform that allows organizations to develop their own AI solutions without vendor lock in, while retaining full control over their training data. The platform supports heterogeneous data sources, model interpretability, usability, and regulatory compliance.

At its core, SA FLaaS is built on globally awarded and peer reviewed federated aggregation and client selection methods, including SimAGG, RegAGG, and RegSimAGG, alongside patent pending innovations. The platform integrates local and global differential privacy mechanisms, encryption techniques, and continuous security monitoring. While primarily targeted at healthcare applications, the technology is designed to be scalable and adaptable to diverse operational environments.

The project focuses on proof of concept validation and the delivery of a fully operational FL platform incorporating pioneering technologies. Ultimately, the project aims to translate research into a product named Secure and Accurate Federated Learning as a Service, the first Finnish made federated learning platform designed to meet stringent privacy and regulatory requirements.

Read more

• SA-FLaaS example animation
• Turku UAS student’s thesis is nominee for AI Thesis of the Year