The global federated learning market size reached USD 112.7 Million in 2021 and is expected to register a revenue CAGR of 10.5% and reach USD 305.7 Million in 2032. Federated Learning (FL) is a decentralized machine learning approach that enables model training across distributed data providers while preserving data privacy. While FL enhances data privacy and security, it faces challenges such as noise-accuracy trade-offs, system and statistical heterogeneity, communication bottlenecks, poisoning attacks, and efficiency versus privacy trade-offs.

The project builds on the PRIVASA project (2021-23), which presented award-winning FL solutions for federated tumor segmentation in brain cancer imaging. The project aims to address FL’s privacy challenges and develop a secure and efficient decentralized framework integrating differential privacy and reinforcement learning. The project´s focus includes data quality, model interpretability, and regulatory compliance in real-world healthcare settings.

The project has filed patents for two key innovations:

1. Differential Privacy (DP) to enhance FL security by adding controlled noise to model updates without compromising output.
2. Dynamic aggregation of model updates, selecting the most promising collaborators automatically.

These innovations align with top-tier privacy-preserving AI standards and are ready for seamless integration into FL development environments. The project´s goal is to advance to technology readiness level 7, focusing on proof-of-concept, developing test benches, and a fully operational FL platform with pioneering technologies.

The project aims to transform research into a product named "Secure and Accurate Federated Learning as a Service," which will be the first-ever Finnish-made privacy-compliant federated learning platform.