Affiliation: Simula Research Laboratory
Contact: miroslav@simula.no
Research interest
Broadly speaking my research falls under the label of Scientific Machine Learning where classical algorithms are combined with more modern tools for machine learning. In the classical category, I am interested in the design, analysis and implementation of efficient numerical schemes for solving multiphysics models with a particular focus on the finite element method and monolithic preconditioning. In the (Sci)ML category, my activities are centered around Hybrid FEM-NN models, which integrate standard FEM discretizations with representations in terms of neural networks in order to encode prior knowledge of the physical constraints. In particular, I am interested in training strategies, stability and approximation properties and applications of the hybrid models.
Short Bio
I obtained a MSc degree in mathematical modeling at Charles University in Prague and PhD in applied mathematics at the University of Oslo. My thesis focused on developing efficient iterative algorithms for coupled multiphysics problems. Following postdoctoral stays at UiO and Simula I am currently a research scientist at Simula Research Laboratory.
Supervisor for the following CompSci projects
-
Mathematics and Statistics
- Stable representations in deep learning (available in call 2)