Gradient-based dimension reduction of multivariate vector-valued functions

DOI10.1137/18M1221837zbMATH Open1433.41007arXiv1801.07922MaRDI QIDQ5220403FDOQ5220403

Authors: Olivier Zahm, P. Constantine, Clémentine Prieur, Youssef M. Marzouk 

Publication date: 20 March 2020

Published in: SIAM Journal on Scientific Computing (Search for Journal in Brave)

Abstract: Multivariate functions encountered in high-dimensional uncertainty quantification problems often vary most strongly along a few dominant directions in the input parameter space. We propose a gradient-based method for detecting these directions and using them to construct ridge approximations of such functions, in the case where the functions are vector-valued (e.g., taking values in $mathbb{R}^n$). The methodology consists of minimizing an upper bound on the approximation error, obtained by subspace Poincar'e inequalities. We provide a thorough mathematical analysis in the case where the parameter space is equipped with a Gaussian probability measure. The resulting method generalizes the notion of active subspaces associated with scalar-valued functions. A numerical illustration shows that using gradients of the function yields effective dimension reduction. We also show how the choice of norm on the codomain of the function has an impact on the function's low-dimensional approximation.

Full work available at URL: https://arxiv.org/abs/1801.07922

Recommendations

• Symmetry results for decay solutions of elliptic systems in the whole space
• Active subspace methods in theory and practice: applications to kriging surfaces
• Gaussian processes with built-in dimensionality reduction: applications to high-dimensional uncertainty propagation
• Gradient free active subspace construction using Morris screening elementary effects
• Dimension reduction via Gaussian ridge functions

zbMATH Keywords

dimension reductionhigh-dimensional function approximationSobol' indicesactive subspaceridge approximationPoincaré inequality

Mathematics Subject Classification ID

Multidimensional problems (41A63) Approximation by other special function classes (41A30) Algorithms for approximation of functions (65D15)

Cites Work

• Principal component analysis.
• Sliced Inverse Regression for Dimension Reduction
• Comment
• Title not available (Why is that?)
• Title not available (Why is that?)
• Sufficient dimension reduction and prediction in regression
• Dimension Reduction for Multivariate Response Data
• Projection pursuit
• Uncertainty management in simulation-optimization of complex systems. Algorithms and applications
• Exploring Regression Structure Using Nonparametric Functional Estimation
• On Principal Hessian Directions for Data Visualization and Dimension Reduction: Another Application of Stein's Lemma
• Title not available (Why is that?)
• Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates
• Theory and practice of finite elements.
• Learning functions of few arbitrary linear parameters in high dimensions
• Sensitivity Analysis in Practice
• Finite Element Error Analysis of Elliptic PDEs with Random Coefficients and Its Application to Multilevel Monte Carlo Methods
• Global sensitivity analysis: The primer
• Karhunen-Loève approximation of random fields by generalized fast multipole methods
• Title not available (Why is that?)
• An inequality for the multivariate normal distribution
• Finite elements for elliptic problems with stochastic coefficients
• Ridge Functions
• Dimensionality reduction and polynomial chaos acceleration of Bayesian inference in inverse problems
• Entropy and sampling numbers of classes of ridge functions
• Derivative based global sensitivity measures and their link with global sensitivity indices
• Sensitivity analysis for multidimensional and functional outputs
• Sensitivity indices for multivariate outputs
• On Nonlinear Functions of Linear Combinations
• Capturing ridge functions in high dimensions from point queries
• Second order Poincaré inequalities and CLTs on Wiener space
• Active Subspaces
• Derivative-Based Global Sensitivity Measures and Their Link with Sobol’ Sensitivity Indices
• Asymptotics for pooled marginal slicing estimator based on SIR\(_\alpha\) approach
• Erratum: Active Subspace Methods in Theory and Practice: Applications to Kriging Surfaces
• An Adaptive Sparse Grid Algorithm for Elliptic PDEs with Lognormal Diffusion Coefficient
• Some extensions of multivariate sliced inverse regression
• On dimension reduction in regressions with multivariate responses
• Certified dimension reduction in nonlinear Bayesian inverse problems
• Derivative-based global sensitivity measures: general links with Sobol' indices and numerical tests
• Poincaré inequalities on intervals -- application to sensitivity analysis
• AN APPROXIMATION THEORETIC PERSPECTIVE OF SOBOL' INDICES WITH DEPENDENT VARIABLES
• Time‐dependent global sensitivity analysis with active subspaces for a lithium ion battery model

Cited In (28)

• Modified Active Subspaces Using the Average of Gradients
• Multi‐fidelity data fusion through parameter space reduction with applications to automotive engineering
• Reduced basis methods for time-dependent problems
• Certified dimension reduction in nonlinear Bayesian inverse problems
• Principal feature detection via \(\phi \)-Sobolev inequalities
• Optimal representation of multivariate functions or data in visualizable low-dimensional spaces
• Embedded ridge approximations
• Derivative-Based Global Sensitivity Analysis for Models with High-Dimensional Inputs and Functional Outputs
• Model Reduction for Nonlinear Systems by Balanced Truncation of State and Gradient Covariance
• Data-free likelihood-informed dimension reduction of Bayesian inverse problems
• A Supervised Learning Approach Involving Active Subspaces for an Efficient Genetic Algorithm in High-Dimensional Optimization Problems
• Preintegration via Active Subspace
• Nonlinear dimension reduction for surrogate modeling using gradient information
• Derivative-informed neural operator: an efficient framework for high-dimensional parametric derivative learning
• Symmetry results for decay solutions of elliptic systems in the whole space
• Prior normalization for certified likelihood-informed subspace detection of Bayesian inverse problems
• Derivative-informed projected neural networks for high-dimensional parametric maps governed by PDEs
• Kernel‐based active subspaces with application to computational fluid dynamics parametric problems using the discontinuous Galerkin method
• Generalized bounds for active subspaces
• Learning high-dimensional parametric maps via reduced basis adaptive residual networks
• Characterization of flow through random media via Karhunen-Loève expansion: an information theory perspective
• A distributed active subspace method for scalable surrogate modeling of function valued outputs
• On the Deep Active-Subspace Method
• Large-scale Bayesian optimal experimental design with derivative-informed projected neural network
• A local approach to parameter space reduction for regression and classification tasks
• Multifidelity Dimension Reduction via Active Subspaces
• An efficient dimension reduction for the Gaussian process emulation of two nested codes with functional outputs
• Structure exploiting methods for fast uncertainty quantification in multiphase flow through heterogeneous media

This page was built for publication: Gradient-based dimension reduction of multivariate vector-valued functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5220403)