Meta modeling

The current section is dedicated to the building of an analytical approximation of the model of a given model. Such an approximation is commonly referred to as a response surface in the literature. This methodology is relevant if each model evaluation is time consuming. Indeed, once a surrogate model has been built up, the various propagation steps may be applied at a negligible cost.

Local approximation

• Taylor expansion

Least squares surrogate models

• Least squares meta models
• Sparse least squares meta model

Gaussian Process Regression

• Gaussian process regression

Functional chaos

• Functional Chaos Expansion
• Multivariate Orthonormal basis
• Univariate orthonormal polynomials
• Tensorized multivariate basis enumeration functions

Surrogate validation

• Validation and cross validation of metamodels