Learn about our research
We are passionate about tackling the latest challenges
in the application of Artificial Intelligence and Machine Learning in the life sciences.
to tackle the lack of training data
In medical imaging, and the classification of either cell types or diseases, there is no ground truth, and labelling data are sparse. This is a unique challenge in biological applications compared to other areas in AI research.
Explainability and uncertainty quantification
the need for insights into models beyond (average) predictive performance
We aim to derive causal and/or mechanistic insights and quantify model uncertainty.
Privacy-aware and federated learning
to preserve privacy while allowing aggregation of information
The ability to deploy AI to relevant problems is linked to the volume of data that can be leveraged for training. This demands federated algorithms and infrastructures that preserve privacy while allowing to aggregate information across different computing centers and country borders.
Sparse predictions from dense inputs
creating sparse summaries of terapixel data sets
Modern imaging yields terapixel data sets, but downstream analytics often require structured, sparse summaries of this data. Current deep learning methods are not able to yield such structured estimates.
Interpretable low-dimensional representations and metric learning
dimensionality reduction and analysis of interrelations
Complex multimodal high-dimensional data (e.g., single-cell multi-omics) require methods for dimensionality reduction and analysis of interrelations, from clusters/classes to continuous gradients to graphs or lineage trees.
Validation and benchmarking for clinical translation
validation and benchmarking of algorithms for specific biomedical projects
Commonly used machine learning metrics and validation strategies often fail to address aspects that are crucial to the application domain and do not allow for tailoring of future research strategies to domain-specific needs.