Embedded ridge approximations
From MaRDI portal
Publication:2020990
Abstract: Many quantities of interest (qois) arising from differential-equation-centric models can be resolved into functions of scalar fields. Examples of such qois include the lift over an airfoil or the displacement of a loaded structure; examples of corresponding fields are the static pressure field in a computational fluid dynamics solution, and the strain field in the finite element elasticity analysis. These scalar fields are evaluated at each node within a discretised computational domain. In certain scenarios, the field at a certain node is only weakly influenced by far-field perturbations; it is likely to be strongly governed by local perturbations, which in turn can be caused by uncertainties in the geometry. One can interpret this as a strong anisotropy of the field with respect to uncertainties in prescribed inputs. We exploit this notion of localised scalar-field influence for approximating global qois, which often are integrals of certain field quantities. We formalise our ideas by assigning ridge approximations for the field at select nodes. This embedded ridge approximation has favorable theoretical properties for approximating a global qoi in terms of the reduced number of computational evaluations required. Parallels are drawn between our proposed approach, active subspaces and vector-valued dimension reduction. Additionally, we study the ridge directions of adjacent nodes and devise algorithms that can recover field quantities at selected nodes, when storing the ridge profiles at a subset of nodes---paving the way for novel reduced order modeling strategies. Our paper offers analytical and simulation-based examples that expose different facets of embedded ridge approximations.
Recommendations
- A near-stationary subspace for ridge approximation
- Gaussian quadrature and polynomial approximation for one-dimensional ridge functions
- Data-driven polynomial ridge approximation using variable projection
- Dimension reduction via Gaussian ridge functions
- Gradient-based dimension reduction of multivariate vector-valued functions
Cites work
- Active subspace methods in theory and practice: applications to kriging surfaces
- Active subspaces. Emerging ideas for dimension reduction in parameter studies
- Adaptive sparse polynomial chaos expansion based on least angle regression
- An Adaptive Estimation of Dimension Reduction Space
- Comment
- Data-driven polynomial ridge approximation using variable projection
- Dimension reduction via Gaussian ridge functions
- Exploring Regression Structure Using Nonparametric Functional Estimation
- Gauss-Christoffel quadrature for inverse regression: applications to computer experiments
- Gradient-based dimension reduction of multivariate vector-valued functions
- scientific article; zbMATH DE number 1220060 (Why is no real title available?)
- scientific article; zbMATH DE number 6159604 (Why is no real title available?)
- Inverse regression for ridge recovery: a data-driven approach for parameter reduction in computer experiments
- Learning functions of few arbitrary linear parameters in high dimensions
- Learning non-parametric basis independent models from point queries via low-rank methods
- Multifidelity Dimension Reduction via Active Subspaces
- Ridge functions
- Separable nonlinear least squares: the variable projection method and its applications
- Sliced Inverse Regression for Dimension Reduction
- Sufficient dimension reduction: methods and applications with R
- User-friendly tail bounds for sums of random matrices
Cited in
(3)
This page was built for publication: Embedded ridge approximations
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2020990)
