Justification for the use of metamodels

Hi everyone,

When we use metamodels in UQ, the metamodels themselves can introduce errors/uncertainties in the results. How do we justify the use of metamodels? Is there any papers specifically on this issue?

Hey Hongzhou,
Maybe you can check: https://www.researchgate.net/publication/317396793_Surrogate_models_for_uncertainty_quantification_An_overview

From my point of view, we validate the surrogate model and quantify the errors within it, this is why for example we measure LOO error, validation error and modified validation errors.
Hope this would help you?

Thank you very much, Tamadur, for the reference. And yes you are right, having a low validation error of the surrogate model is a good justification of implementing it.



1 Like