Local inversion-free estimation of spatial Gaussian processes

DOI10.1214/19-EJS1592zbMATH Open1435.62349arXiv1811.12602OpenAlexW2982170000MaRDI QIDQ2008608FDOQ2008608

Authors: Hossein Keshavarz, Xuanlong Nguyen, Clayton Scott 

Publication date: 26 November 2019

Published in: Electronic Journal of Statistics (Search for Journal in Brave)

Abstract: Maximizing the likelihood has been widely used for estimating the unknown covariance parameters of spatial Gaussian processes. However, evaluating and optimizing the likelihood function can be computationally intractable, particularly for large number of (possibly) irregularly spaced observations, due to the need to handle the inverse of ill-conditioned and large covariance matrices. Extending the "inversion-free" method of Anitescu, Chen and Stein cite{anitescu2017inversion}, we investigate a broad class of covariance parameter estimators based on inversion-free surrogate losses and block diagonal approximation schemes of the covariance structure. This class of estimators yields a spectrum for negotiating the trade-off between statistical accuracy and computational cost. We present fixed-domain asymptotic properties of our proposed method, establishing $sqrtn$-consistency and asymptotic normality results for isotropic Matern Gaussian processes observed on a multi-dimensional and irregular lattice. Simulation studies are also presented for assessing the scalability and statistical efficiency of the proposed algorithm for large data sets.

Full work available at URL: https://arxiv.org/abs/1811.12602

Recommendations

• On the consistency of inversion-free parameter estimation for Gaussian random fields
• Covariance tapering for likelihood-based estimation in large spatial data sets
• Asymptotic analysis of the role of spatial sampling for covariance parameter estimation of Gaussian processes
• Covariance tapering for multivariate Gaussian random fields estimation
• Estimating spatial covariance using penalised likelihood with weightedL₁penalty

zbMATH Keywords

Gaussian processcomputationally scalablefixed-domain asymptotic analysisirregularly spaced observationslocal inversion-free covariance estimation

Mathematics Subject Classification ID

Inference from spatial processes (62M30) Random fields; image analysis (62M40) Gaussian processes (60G15) Random fields (60G60) Non-Markovian processes: estimation (62M09)

Cites Work

• On fixed-domain asymptotics and covariance tapering in Gaussian random field models
• Approximating Likelihoods for Large Spatial Data Sets
• A Limited Memory Algorithm for Bound Constrained Optimization
• The role of the range parameter for estimation and prediction in geostatistics
• Covariance tapering for likelihood-based estimation in large spatial data sets
• Inconsistent Estimation and Asymptotically Equal Interpolations in Model-Based Geostatistics
• Handbook of spatial statistics.
• Hanson-Wright inequality and sub-Gaussian concentration
• Optimal change point detection in Gaussian processes
• On the consistent separation of scale and variance for Gaussian random fields
• Stochastic approximation of score functions for Gaussian processes
• Difference filter preconditioning for large covariance matrices
• Asymptotic properties of a maximum likelihood estimator with data from a Gaussian process
• On the use of discrete Laplace operator for preconditioning kernel matrices
• On the consistency of inversion-free parameter estimation for Gaussian random fields
• A matrix-free approach for solving the parametric Gaussian process maximum likelihood problem

Cited In (9)

• Scalable Physics-Based Maximum Likelihood Estimation Using Hierarchical Matrices
• Estimating the parameters of some common Gaussian random fields with nugget under fixed-domain asymptotics
• On the consistency of inversion-free parameter estimation for Gaussian random fields
• Local solutions to inverse problems in geodesy. The impact of the noise covariance structure upon the accuracy of estimation
• Asymptotic behavior of the likelihood function of covariance matrices of spatial Gaussian processes
• On fixed-domain asymptotics, parameter estimation and isotropic Gaussian random fields with Matérn covariance functions
• Fitting Matérn smoothness parameters using automatic differentiation
• Asymptotic analysis of the role of spatial sampling for covariance parameter estimation of Gaussian processes
• Fast spatial Gaussian process maximum likelihood estimation via skeletonization factorizations

Uses Software

• LBFGS-B

This page was built for publication: Local inversion-free estimation of spatial Gaussian processes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2008608)