Federated Learning for Healthcare Research Collaboration
Federated learning (FL) has emerged as a promising approach for modelling real-world problems using decentralised and heterogeneous data while preserving data privacy. However, our understanding of how to develop robust, principled, and generalisable federated learning models for imaging tasks remains limited, particularly in the small and imbalanced data settings that are common in medical imaging.
This work package aims to establish the scientific foundations for principled FL-based solutions for supervised and unsupervised image and video analysis, as well as multimodal imaging applications. The goal is to enable seamless modelling of imaging problems using decentralised and heterogeneous datasets, achieving performance comparable to, or approaching, that obtained when all data are available within a single centre. At the same time, the research seeks to preserve data privacy and to better understand the fundamental limits and conditions under which such performance can be achieved.