Hierarchical Reduced Order Models in Bayesian Inversion augmented by Deep Learning
Ongoing work on improving reduced order models by better characterizing their error using deep learning methods.
Currently implemented work characterizes the error of a reduced order model solving the heat conduction problem in a thermal fin. The problem statement is defined in Prof. Tan Bui's thesis.
A high fidelity finite element solver is implemented using FEniCS.
A reduced basis is formed using a greedy model-constrained adaptive sampling method suggested by Prof. Bui.
A Tensorflow estimator is used to learn the error between the high fidelity model and the reduced model.
- FEniCS 2018.1
- hippylib 2.1.0
- python 3.6.6
- tensorflow 1.11.0
- pandas 0.23.4