The ELLIS unit Heidelberg will foster innovations at the interface of artificial intelligence (AI), machine learning (ML) and the biological and medical sciences. The mission of the unit is to facilitate breakthrough applications of AI/ML, delivering leading-edge analytics to fully exploit the rapidly growing volumes of biomedical data across Europe. The unit will conduct foundational research to address key challenges and obstacles for deploying AI in biomedicine. This includes methods to cope with the heterogeneous and often noisy nature of “omics” data and the scarcity of labeled data in medical imaging, algorithms and infrastructures to deal with ethical and privacy constraints of data access, algorithms to infer causal relationships, as well as novel modelling strategies to deliver interpretable, auditable decisions. The research will be carried out with a particular focus on (1) un-/semi-/self-supervised learning to tackle lack of training data, (2) explainability and uncertainty quantification, (3) privacy aware and federated learning, (4) sparse predictions from dense inputs and (5) interpretable low-dimensional representations and metric learning. Research activities of the unit will be complemented with a pan ELLIS service mission, democratizing access to curated reference datasets, as well as organizing and certifying benchmarks of novel AI-based analytics.

Unit Director