search

katiana22 / surrogate-overparameterization

Source code of "On the influence of over-parameterization in manifold based surrogates and deep neural operators".
MIT License
5 stars 0 forks source link

Availability of functions for generating initial conditions? #1

Closed newalexander closed 11 months ago

newalexander commented 1 year ago

Hello, and thanks for making this public; the paper is very interesting. For more general usability and extension of these results, would it be possible to upload the code you used to generate the initial conditions for the Brusselator solutions?

In addition, did you experiment with any spatial discretizations other than the 28 x 28 used in the paper?

katiana22 commented 12 months ago

Hello,

Thank you for your interest in our code repository and I apologize for the late reply!

You can generate the initial conditions using UQpy's 2D KarhunenLoeveExpansion2D class using as correlation function a squared exponential with the length scale parameter values provided in the manuscript: https://uqpyproject.readthedocs.io/en/latest/auto_examples/stochastic_processes/karhunen_loeve_2d/plot_karhunen_loeve_2d.html#sphx-glr-auto-examples-stochastic-processes-karhunen-loeve-2d-plot-karhunen-loeve-2d-py

Regarding your second question - we have not analyzed the sensitivity of the models to the PDE discretization since the 28x28 grid is sufficient for the studied problem, but it would be interesting to explore more complex applications that require more fine discretizations and high fidelity simulations.