Chair of Explainable Machine Learning
The Chair of Explainable Machine Learning (xAILab Bamberg) develops machine learning methods that remain reliable outside the lab, with medical image analysis as the main application. Our work is for clinicians and companies who need models they can check, for students who want to learn deep learning, and for researchers who reuse our open code and benchmarks.
The chair belongs to the Faculty of Information Systems and Applied Computer Sciences of the University of Bamberg and has been led by Professor Christian Ledig since April 2022. We teach deep learning and the mathematics of machine learning, with seminars, projects and theses, and the code, benchmarks and datasets of our papers are openly available.
What we work on
- Models that keep working when the data change: training methods that hold up across hospitals, scanners and patient groups.
- Knowing when not to trust a model: detectors that flag unfamiliar images before a prediction is used.
- Learning from data that cannot be shared or fully trusted: privacy-preserving data sharing and learning with wrong labels.
- Testing models under realistic conditions: benchmarks and protocols that are closer to clinical practice.
For students
Our courses and open thesis topics.
For applicants
Open positions for PhD candidates and postdocs.
For collaborators and press
Our publications, software and datasets and contact details.
Recent publications
Three of our latest papers. All publications by year are on the page Publications.
- Meike Bauer, Marco Lents, Erik Schmidt, Tim Hamann, Lukas Pieger, Susanne Salata, Francesco Di Salvo, Tal Hoffmann, Jens Barth, Christian Ledig: Ranking handwriting legibility with foundation-model embeddings at human-level consistency on the HaLO dataset. Scientific Reports 16, 31913 (2026). Paper · Code · Data
- Francesco Di Salvo, Sebastian Doerrich, Christian Ledig: MaRS: Robust Out-of-Distribution Detection via Mahalanobis Residual Scoring. MICCAI 2026. Paper · arXiv · Code
- Sebastian Doerrich, Francesco Di Salvo, Shyam Nandan Rai, Marco Lents, Christian Ledig: Simple, Safe, and Overlooked: Reclaiming Sustainable Domain Generalization with Statistical Color Matching. MICCAI 2026 Workshop DEMI. Best Paper Award Runner-Up. Paper · Code
![[Translate to English:] [Translate to English:]](/fileadmin/_processed_/c/0/csm_xaiLogo_halftransparent_e3541bc9d6.png)
