Finite element representations of Gaussian processes: balancing numerical and statistical accuracy

DOI10.1137/21M144788XMaRDI QIDQ5052906zbMATH OpenOpenAlexFDO

Authors Daniel Sanz-Alonso, Ruiyi Yang

Publication date 25 November 2022

Published in SIAM/ASA Journal on Uncertainty Quantification (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/2109.02777

zbMATH Keywords

Bayesian nonparametricsfinite elementsMatérn Gaussian processes

Mathematics Subject Classification ID

Computational methods for problems pertaining to statistics (62-08) Asymptotic properties of nonparametric inference (62G20)

Abstract: The stochastic partial differential equation approach to Gaussian processes (GPs) represents Mat'ern GP priors in terms of $n$ finite element basis functions and Gaussian coefficients with sparse precision matrix. Such representations enhance the scalability of GP regression and classification to datasets of large size $N$ by setting $napproxN$ and exploiting sparsity. In this paper we reconsider the standard choice $napproxN$ through an analysis of the estimation performance. Our theory implies that, under certain smoothness assumptions, one can reduce the computation and memory cost without hindering the estimation accuracy by setting $nllN$ in the large $N$ asymptotics. Numerical experiments illustrate the applicability of our theory and the effect of the prior lengthscale in the pre-asymptotic regime.

Recommendations

• Variational Fourier features for Gaussian processes
• Convergence of sparse variational inference in Gaussian processes regression
• Efficient Gaussian process regression for large datasets
• Sparse on-line Gaussian processes
• Sparse spectrum Gaussian process regression

Cites work

• A case study competition among methods for analyzing large spatial data
• A comparison between Markov approximations and other methods for large spatial data sets
• A general framework for Vecchia approximations of Gaussian processes
• A simple condition for asymptotic optimality of linear predictions of random fields
• A unifying view of sparse approximate Gaussian process regression
• An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach
• Analysis of Boundary Effects on PDE-Based Sampling of Whittle--Matérn Random Fields
• Approximation, metric entropy and small ball estimates for Gaussian measures
• Bounds on the efficiency of linear predictions using an incorrect covariance function
• Consistency of Bayesian inference with Gaussian process priors in an elliptic inverse problem
• Convergence Rates for Penalized Least Squares Estimators in PDE Constrained Regression Problems
• Convergence rates of posterior distributions.
• Correlation-based sparse inverse Cholesky factorization for fast Gaussian-process inference
• Dimension-independent likelihood-informed MCMC
• Fixed-domain asymptotic properties of tapered maximum likelihood estimators
• Gaussian Markov Random Fields
• Gaussian processes for machine learning.
• Hilbert space methods for reduced-rank Gaussian process regression
• scientific article; zbMATH DE number 3555381 (Why is no real title available?)
• scientific article; zbMATH DE number 1149759 (Why is no real title available?)
• scientific article; zbMATH DE number 1522714 (Why is no real title available?)
• scientific article; zbMATH DE number 967584 (Why is no real title available?)
• scientific article; zbMATH DE number 962825 (Why is no real title available?)
• Information rates of nonparametric Gaussian process methods
• Interpolation of spatial data. Some theory for kriging
• Introduction to uncertainty quantification
• Inverse Problems and Data Assimilation
• Inverse problems: a Bayesian perspective
• MCMC methods for functions: modifying old algorithms to make them faster
• Numerical solution of fractional elliptic stochastic PDEs with spatial white noise
• On fixed-domain asymptotics and covariance tapering in Gaussian random field models
• ON STATIONARY PROCESSES IN THE PLANE
• On the consistency of graph-based Bayesian semi-supervised learning and the scalability of sampling algorithms
• Operator-adapted wavelets, fast solvers, and numerical homogenization. From a game theoretic approach to numerical approximation and algorithm design
• Predicting random fields with increasing dense observations
• Pushing the Limits of Contemporary Statistics: Contributions in Honor of Jayanta K. Ghosh
• Rates of contraction of posterior distributions based on Gaussian process priors
• Regularity and convergence analysis in Sobolev and Hölder spaces for generalized Whittle-Matérn fields
• Sparse Cholesky Factorization by Kullback--Leibler Minimization
• Spatio-temporal modeling of particulate matter concentration through the SPDE approach
• Statistical Numerical Approximation
• Studies in the history of probability and statistics XLIX: On the Matérn correlation family
• The Rational SPDE Approach for Gaussian Random Fields With General Smoothness
• The screening effect in kriging
• The SPDE approach to Matérn fields: graph representations
• The vectorization of ITPACK 2C
• Thomas Bayes' walk on manifolds
• Uniform asymptotic optimality of linear predictions of a random field using an incorrect second-order structure

Cited in

(13)

• The statistical finite element method (statFEM) for coherent synthesis of observation data and model predictions
• INFERENCE AND UNCERTAINTY PROPAGATION OF ATOMISTICALLY INFORMED CONTINUUM CONSTITUTIVE LAWS, PART 2: GENERALIZED CONTINUUM MODELS BASED ON GAUSSIAN PROCESSES
• Stochastic PDE representation of random fields for large-scale Gaussian process regression and statistical finite element analysis
• Theoretical Guarantees for the Statistical Finite Element Method
• The Matérn model: a journey through statistics, numerical analysis and machine learning
• Elliptic Bayesian inverse problems on metric graphs
• Multiple and weak Markov properties in Hilbert spaces with applications to fractional stochastic evolution equations
• Equispaced Fourier representations for efficient Gaussian process regression from a billion data points
• Gaussian process regression under computational and epistemic misspecification
• Sparse Cholesky factorization for solving nonlinear PDEs via Gaussian processes
• Bayesian optimization on networks
• Mass Lumping and Numerical Quadrature for Approximation of Fractional Elliptic Differential Equations Driven by Gaussian White Noise
• Asymptotics of numerical integration for two-level mixed models

Describes a project that uses

Uses Software

• GMRFLib
• Stem

This page was built for publication: Finite element representations of Gaussian processes: balancing numerical and statistical accuracy

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5052906)