How do I learn physical models from data? Can I trust generative models in medical imaging? How to reconstruct magnetic resonance images from noisy, incomplete data?

Questions like these are addressed in our research group, which works at the interface of mathematics, computer science and interdisciplinary applications. In previous works, we have developed methods to predict error bound for generative models in imaging, to learn parts of physical models from data or to [reconstruct dynamic magnetic resonance images of the beating heart. We have further worked on interdisciplinary applications in electron tomography, emergency care medicine or positron emission tomography.

Our current methodological research focuses on generative models in machine learning, inverse problems for partial differential equations with learned components and variational methods in imaging. The contributions of our work are both of mathematical nature, comprising mathematical analysis and novel machine learning methods, and strongly application-driven, comprising the implementation and publication of parallelized algorithms and software tools for dealing with real-world data. As part of our research, we are always interested in opening up new, interdisciplinary fields of application together with cooperation partners and in using machine learning methods to generate a positive impact on science, technology and society.

Further information and all publications of our group can be found on the website of the head of our research group.