Structure-Preserving Neural Networks for Scientific Computing
Abstract
In this dissertation we propose an array of novel neural network architectures and optimization algorithms to treat problems in scientific computing, especially reduced order modeling. All the new architectures are structure-preserving (symplectic, volume-preserving, …) in some sense and we show how these structure-preserving properties can improve the application of neural networks to problems from scientific computing.
BibTeX
@phdthesis{dissertation,
author = {Brantner, Benedikt},
title = {Geometric Machine Learning},
year = {2025},
school = {Technische Universität München},
pages = {306},
language = {en},
abstract = {In this dissertation we propose an array of novel neural network architectures and optimization algorithms to treat problems in scientific computing, especially reduced order modeling. All the new architectures are structure-preserving (symplectic, volume-preserving, ...) in some sense and we show how these structure-preserving properties can improve the application of neural networks to problems from scientific computing.},
keywords = {},
note = {},
url = {https://mediatum.ub.tum.de/1754855},
doi = {},
}