Inverse Uncertainty Quantification of Compute-intensive Models Using Machine Learning Methods
Inverse problems is hard to solve when the model is compute-intensive. In this study, I use machine leaning method to solve an inverse problem.
Autoencode is trained to learn from a small amount of simulation data, resulting in a lower dimensional representation of the full state of the simulation.
Exploring the lower dimensional latent space and reconstructing the model state using the decoder, the parameters of the model are easily found.