Federated Learning

Multi-centre evidence without moving clinical data — collaborative, privacy-preserving learning across hospitals.

Clinical data are scattered across hospitals and cannot simply be pooled. With federated learning we train and evaluate models across institutions while the data stay where they are — handling heterogeneous (non-IID) data, label scarcity, personalisation and fairness between sites. Through OMI and the Medical Informatics Initiative we bring these methods to remote AI inference across German university hospitals.

Key questions

  • How can hospitals train a shared model when their data are non-IID and cannot leave the premises?
  • How do we personalise a federated model for each site while keeping it fair to all?
  • How do we generate multi-centre clinical evidence without moving patient data?

Research Collaboration and Funded Projects 11 filter all →

Publications 23 filter all →

All 23 publications →

Datasets 3

Code 5

Teaching 8

In the media 3

Community & events 58