Federated Learning ARDC

The ARDC Frontier Federated Machine Learning Capacity Building for Australia project is a national initiative aimed at advancing privacy-preserving artificial intelligence and machine learning capabilities across Australia’s healthcare and research sectors. Running from 2025 to 2027, the project focuses on enabling hospitals, registries and research centres to collaboratively develop machine learning models without sharing sensitive patient data. By leveraging Federated Learning, the project seeks to overcome major privacy, governance and data-sharing barriers while supporting scalable, secure and trusted AI-driven healthcare research.

The project is structured into six interconnected Working Packages (WPs) that collectively address the technical, operational and governance requirements for federated learning adoption in Australia. These include program coordination and community development (WP1), federated learning platforms and models (WP2), secure deployment and interoperability (WP3), virtual research environment architecture (WP4), governance and standards frameworks (WP5) and training resources for FL adoption and analysis (WP6) (Fig. 1). Together, these work packages establish the infrastructure, policies, expertise and collaborative ecosystem required to support real-world federated learning applications across Australian institutions. At UNSW, our primary contribution is within WP2 – Federated Learning Platforms and Models, undertaken in collaboration with the University of Sydney. Our work focuses on evaluating and benchmarking open-source federated learning tools, implementing horizontal and vertical federated learning models and developing frameworks to address data imbalance and reconstruction risks. We are currently deploying and testing federated learning workflows using platforms such as FLOWER on both local systems and the Nectar cloud, with planned demonstrations using distributed radiotherapy datasets across Australian healthcare networks. This work aims to support secure, scalable and production-ready federated learning solutions for sensitive healthcare data applications.