In this talk, I will present recent work on estimating kinetic properties of molecular systems - such as transition timescales and correlation functions - using models for the Koopman generator. I will show that random Fourier features - a low-rank approximation technique for kernel methods - provide a versatile and efficient framework to estimate these models from data. I will present three use cases: first, a benchmark study on estimating slow transition timescales. Second, interpolation of kinetic properties across temperatures using generative models. Third, learning of coarse grained models which preserve transition timescales.
Feliks Nüske received his Ph.D. in Mathematics from Freie Universität Berlin, Germany, in 2017. He subsequently held postdoctoral positions as a Rice Academy Fellow at Rice University (U.S.), as well as at Universität Paderborn (Germany). Since 2022, he has been an independent Max-Planck Research Group Leader at the MPI for Dynamics of Complex Technical Systems in Magdeburg, Germany. His research interests include modeling and analysis of stochastic dynamics, machine learning for dynamical systems, as well as modeling and simulation of molecular systems.