Our group develops methods that make machine learning faster, more reliable, and easier to use. The computational methods that drive learning machines are typically numerical algorithms: optimization methods (e.g. to train deep networks), solvers for ordinary differential equations (e.g. for diffusion models, reinforcement learning and control) and partial differential equations (for scientific modelling from massive data), and large-scale linear algebra as the base layer of much higher-level functionality (like uncertainty quantification for deep learning, fine-tuning, the inclusion of physical knowledge in data-centric models, and much else). 

We have a history of contributing to the mathematical and even philosophical foundations of computation, but also the development of concrete software artefacts. Together with international collaborators, we helped establish the field of probabilistic numerics. An example (out of many) of our technical contributions to this field include the general formalism of ODE Filters. 

As AI and machine learning is now affecting all domains of science, our group increasingly tries to contribute not just to the foundations of the field, but also to concrete applications that pose deep technical challenges. 

The overwhelming majority of our research is funded by public funds. In the past, we benefited from support by the Emmy Noether Program of the DFG; an independent group grant of the Max Planck Society; and two ERC (Starting and Consolidator) grants.