- Brendan Keith, Lawrence Livermore National Laboratory
- Maciej Paszynski, AGH University of Science and Technology
Finite element methods are peerless in their approximation power, scalability, and applicability in countless scientific and engineering applications. In turn, decades of research have established a vast body of knowledge related to the discretization of PDEs with such methods. In practice, specific discretization decisions are guided by this vast domain knowledge but are still rarely optimal. Recent work has shown that these decisions may be learned in both supervised and unsupervised ways, thus resulting in new and general paradigms for optimizing finite element approximations.
This minisymposium focuses on showcasing new tools and techniques which can improve the approximation power of finite element methods. Examples include (but are not limited to) data-driven adaptive mesh refinement/unrefinement, order elevation, mesh generation, mesh movement, and remeshing techniques, as well as data-driven stabilization methods.