Frontier Federated Machine Learning Capacity Building for Australia

The Challenge

High-quality health data is collected every day in hospitals, clinics and registries across Australia, but most of it stays locked within individual organisations. Privacy, ethics, governance and technical constraints make it difficult, and often impossible, to pool sensitive data in one place for analysis.

If researchers can only analyse small local datasets, the results may be incomplete and may not apply well to other groups.

Federated learning offers a way forward by training models where the data already lives, but it is still an emerging area. Australia lacks shared infrastructure, agreed architectures, clear governance pathways and practical guidance for setting up new federated learning networks in health. Projects are often bespoke, difficult to scale and challenging for new teams to join.

The Response

The Federated Machine Learning Network (the Frontier Federated Machine Learning Capacity Building for Australia Project) is creating a shared foundation for federated learning in Australian health research. It brings together 3 established networks:

• Australian Cancer Data Network, University of New South Wales (UNSW)
• Australian Imaging Service, University of Sydney (USyd)
• National Infrastructure for federated learNing in DigitAl health (NINA), The University of Queensland (UQ)
• along with ARDC and QCIF Digital Research to design nationally useful approaches, tools and guidance.

The project is organised into six work packages that together:

• build a national community of practice by establishing an ARDC Federated Learning Interest Group, regular seminars, interactive events and a sustainability plan to support ongoing collaboration
• expand technical capability through recommended platforms, algorithms for combined vertical and horizontal federated learning and methods to address data imbalance and reconstruction risks, with demonstrations on the ARDC Nectar Research Cloud
• streamline secure deployment by defining a maturity model and requirements checklist for trusted federated learning software and testing deployments that support multiple federations on shared hardware
• design a secure reference architecture with national scope for secure federated learning
• create governance and standards frameworks through consultation with health departments, hospitals, registries, ethics committees and consumers to produce templates and guidance aligned with the Five Safes
• deliver training and support with materials and sessions for early adopters.

Together, these activities aim to make it simpler and safer for research teams to decide when federated learning is appropriate, choose suitable tools, navigate governance and participate in collaborative studies.

Who Will Benefit

The project is designed to support several groups:

• health and medical researchers who will gain guidance, tools, platforms and opportunities to connect and learn through the Federated Learning Interest Group
• data custodians and governance teams who will benefit from clearer processes, risk assessment frameworks and governance templates that support compliant use of federated learning
• students and emerging researchers who will have access to new training opportunities
• industry and innovators who will be able to explore federated learning approaches while keeping sensitive data within the control of original custodians.

The Partners

The Federated Machine Learning Network is a co-investment partnership between the ARDC and:

• University of New South Wales (UNSW) – Lead organisation hosting the Australian Cancer Data Network and providing technical and project leadership
• University of Sydney (USyd) – Home of the Australian Imaging Service, contributing imaging, federated learning and secure deployment expertise
• University of Queensland (UQ) – Lead node for NINA, contributing governance, health informatics and sociotechnical expertise
• QCIF Digital Research  – Providing cloud, HPC, AI, governance and training expertise and leading reference architecture and training work packages.

The project is also engaging health departments, hospitals, registries, ethics committees and consumers through targeted consultations and workshops.

Target Outcomes

The project aims to deliver:

• a functioning Federated Learning Interest Group with a sustainability plan and regular events
• recommendations for selecting federated learning platforms and published guidance for the Australian context
• frameworks, algorithms and demonstrations for combined vertical and horizontal federated learning, data imbalance mitigation and reconstruction risk assessment
• a maturity model and requirements checklist for secure federated learning software deployment
• documentation for using shared hardware across multiple federated networks
• a reference architecture for secure federated learning and analysis on ARDC Nectar Research Cloud
• consultation reports, risk templates and unified governance frameworks aligned with the Five Safes
• training materials and recorded sessions for early adopters in health research organisations.

Key Resources

• ARDC Nectar Research Cloud
• Australian Cancer Data Network
• Australian Imaging Service
• National Infrastructure for Federated Learning in Digital Health (NINA)