Research

Spectral analysis in form of the eigendecomposition of the Laplace-Beltrami operator is considered the Swiss-Army knife of shape analysis. It is widely used for the non-rigid shape correspondence problem because it allows to make the problem low-dimensional and continuous which is especially valuable in deep learning settings. However, spectral properties are sensitive to surface noise, like holes, topological changes and inaccurate vertex place as can often happen in 3D scanning, and properties from theory often do not hold in practice. The goal of this project to make shape correspondence methods more robust against this kind of noise by explicitly and implicitly modeling the noise sources during optimization.

Plasma fusion has the potential to revolutionize clean and safe energy. One promising design is the form of the stellarator in which the plasma is magnetically confined in a torus-like flow. The shape of the flow is essential for the physical and engineering aspects of the fusion reactor. In this project we will combine our expertise in plasma physics, machine learning, and high-performance computing to tackle the automatic optimization of critical components of stellarators. This is a collaboration with the start-up Proxima Fusion, Technical University of Munich and Forschungszentrum Jülich.

We are part of the excellence cluster proposal "Dynaverse" which aims to disentangle the dynamic evolution of our Universe by understanding and finding the connection between the highly nonlinear physical processes that occur on vastly different timescales, from fractions of a second to billions of years. It brings together world-class hub of expertise in radio astronomy, lab-experiments, simulations, and machine learning / artificial intelligence. Astrophysics is entering the era of big data and one goal is to establish expertise in Astroinformatics which we are excited to contribute to. (decision coming 2025)