Federated analytics - A New Paradigm for Data Processing and Machine Learning - Insights from the NINA Project

Federated analytics represents a paradigm shift in the field of data processing and machine learning, offering a decentralised approach that prioritises privacy and control over data. Each participating site independently processes its locally stored data to answer database queries, generate descriptive statistics, or train machine learning models. The key advantage is that these sites only share the results of their computations which are then aggregated to derive global insights or form a global model. As AI-driven research becomes increasingly prevalent, federated learning is garnering significant attention. While federated learning offers significant benefits, it also introduces unique challenges. These challenges include the need for specific skills and robust infrastructure at each participating site, navigating uncharted territory for data governance, and prompting a shift in the governance model of trusted research environments. Addressing these challenges is crucial for the successful implementation and adoption of federated analytics and requires a collaborative approach between servers, platforms, and sites. The National Infrastructure for Federated Learning in Digital Health to Generate New Models of Care for Chronic Diseases (NINA) project is a pioneering initiative in the field of digital health research. It leverages federated learning to address the critical unmet need of chronic diseases. With over 20 partners across Australia, NINA aims to establish a national capability and infrastructure network for enabling federated learning in Australia. This presentation provides an overview of the NINA project, discuss the benefits and challenges of federated learning, and present the strategies adopted by NINA to overcome these challenges.