Properties and comparison of some kriging sub-model aggregation methods

DOI10.1007/S11004-021-09986-2zbMATH Open1495.86009arXiv1707.05708OpenAlexW3134314433MaRDI QIDQ2676512FDOQ2676512

Authors: François Bachoc, Nicolas Durrande, Clément Chevalier, Didier Rullière 

Publication date: 27 September 2022

Published in: Mathematical Geosciences (Search for Journal in Brave)

Abstract: Kriging is a widely employed technique, in particular for computer experiments, in machine learning or in geostatistics. An important challenge for Kriging is the computational burden when the data set is large. This article focuses on a class of methods aiming at decreasing this computational cost, consisting in aggregating Kriging predictors based on smaller data subsets. It proves that aggregation methods that ignore the covariancebetween sub-models can yield an inconsistent final Kriging prediction. In contrast, a theoretical study of the nested Kriging method shows additional attractive properties for it: First, this predictor is consistent, second it can be interpreted as an exact conditional distribution for a modified process and third, the conditional covariances given the observations can be computed efficiently. This article also includes a theoretical and numerical analysis of how the assignment of the observation points to the sub-models can affect the prediction ability of the aggregated model. Finally, the nested Kriging method is extended to measurement errors and to universal Kriging.

Full work available at URL: https://arxiv.org/abs/1707.05708

Recommendations

• Nested kriging predictions for datasets with a large number of observations
• Penalized blind kriging in computer experiments
• Subset selection from large datasets for kriging modeling
• Kriging metamodeling in simulation: a review
• A note on the choice and the estimation of kriging models for the analysis of deterministic computer experiments

zbMATH Keywords

consistencyGaussian processeserror boundsmodel aggregationnested pointwise aggregation of experts

Mathematics Subject Classification ID

Directional data; spatial statistics (62H11) Statistical aspects of big data and data science (62R07) Geostatistics (86A32) Computational methods for problems pertaining to geophysics (86-08)

Cites Work

• Efficient global optimization of expensive black-box functions
• Gaussian Markov Random Fields
• Fixed Rank Kriging for Very Large Spatial Data Sets
• Geostatistics. Modeling spatial uncertainty.
• A case study competition among methods for analyzing large spatial data
• Gaussian processes for machine learning.
• Gaussian Predictive Process Models for Large Spatial Data Sets
• Covariance tapering for likelihood-based estimation in large spatial data sets
• A unifying view of sparse approximate Gaussian process regression
• Improving the performance of predictive process modeling for large datasets
• Convergence properties of the expected improvement algorithm with fixed mean and covariance functions
• Statistics for spatial data
• Nested kriging predictions for datasets with a large number of observations
• The origins of kriging
• Training Products of Experts by Minimizing Contrastive Divergence
• Cross validation and maximum likelihood estimations of hyper-parameters of Gaussian processes with model misspecification
• Asymptotic properties of a maximum likelihood estimator with data from a Gaussian process
• Probability aggregation methods in geoscience
• On the effect of covariance function estimation on the accuracy of kriging predictors
• Cross-validation estimation of covariance parameters under fixed-domain asymptotics
• Development and evaluation of geostatistical methods for non-Euclidean-based spatial covariance matrices

Cited In (4)

• Penalized blind kriging in computer experiments
• Nested kriging predictions for datasets with a large number of observations
• Subset selection from large datasets for kriging modeling
• Gaussian processes for computer experiments

Uses Software

• GMRFLib
• DiceKriging
• DiceOptim
• FRK

This page was built for publication: Properties and comparison of some kriging sub-model aggregation methods

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2676512)