Research Group Uncertainty Quantification
Exploiting mathematical models for simulation and forecasting requires not only sufficiently precise and affordable numerical solvers, but also the exact knowledge of all (important) model parameters. Particularly the latter is not always a given in practice. Natural variations of materials and their properties as well as random perturbations and forces call for a probabilistic approach to treat the limited knowledge about model coefficients. The resulting prediction should then sensibly quantify the corresponding range of plausible outputs.
This is exactly the focus of our research. We develop and analyze efficient methods for the propagation of uncertainty in complex computational models. To this end, we exploit, for instance, tools from high-dimensional approximation such as sparse grid collocation or deep neural networks. Moreover, we work on Bayesian approaches for inverse problems and data assimilation for differential equations, in particular, efficient and robust sampling methods for high-dimensional problems (e. g., Markov chain Monte Carlo, particle methods).
Telefon: +49 3731 39−2798
Fax: +49 3731 39−3442
Email: Katja [dot] Hetzemath [dot] tu-freiberg [dot] de
Technische Universität Bergakademie Freiberg
Fakultät für Mathematik und Informatik