Our Research

Building Methods for Machine Learning

We aim to make machine learning more versatile, efficient and reliable. To this end, our group develops algorithms for the foundations, the inner loop of machine learning. With international and local collaborators, we also develop custom solutions for pressing applications of machine learning in the sciences. 

If you are interested in how we fund our research, you can find a detailed disclosure here

Making better use of data

In collaborations with key academic partners at home and elsewhere, we have developed custom solutions for inference problems in the sciences. Science is a key application domain for machine learning, but ML is now largely driven by industrial interests, and the demands of scientific research do not always match those of industrial use cases. Scientific ML solutions have to leverage the large and structured prior knowledge available in science and return equally structured models as output. And they have to use, express and manage explicit uncertainty in their statements. In our collaborations we leverage our knowledge of the foundations of ML computation to help our partners build efficient, specialised scientific inference tool chains.