Over the last years, different purely data-driven methods for the approximation of the governing equations of dynamical systems or associated transfer operators have been developed. The main advantage is that these methods can be applied to measurement or simulation data, knowledge about the underlying system itself is not required.
The goal of this seminar is to study different machine learning methods for dynamical systems, of particular interest are molecular dynamics and fluid dynamics applications as well as text and video data analysis.
Topics of interest include:
- model reduction
- learning governing equations from data
- transfer operator approximation
- computation of metastable and coherent sets
- reproducing kernel Hilbert spaces and kernel mean embeddings