Posterior contraction in Gaussian process regression using Wasserstein approximations

DOI10.1093/IMAIAI/IAX003MaRDI QIDQ4603714zbMATHOpenAlexFDO

Authors Anirban Bhattacharya, Debdeep Pati

Publication date 19 February 2018

Published in Information and Inference: A Journal of the IMA (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1502.02336

zbMATH Keywords

kernel regressionGaussian processWasserstein distanceposterior contractionHermite polynomialsrandom design

Mathematics Subject Classification ID

Nonparametric regression and quantile regression (62G08) Asymptotic properties of nonparametric inference (62G20) Gaussian processes (60G15) Signal theory (characterization, reconstruction, filtering, etc.) (94A12)

Abstract: We study posterior rates of contraction in Gaussian process regression with unbounded covariate domain. Our argument relies on developing a Gaussian approximation to the posterior of the leading coefficients of a Karhunen--Lo'{e}ve expansion of the Gaussian process. The salient feature of our result is deriving such an approximation in the $L^2$ Wasserstein distance and relating the speed of the approximation to the posterior contraction rate using a coupling argument. Specific illustrations are provided for the Gaussian or squared-exponential covariance kernel.

Recommendations

• Rates of contraction of posterior distributions based on Gaussian process priors
• Lower bounds for posterior rates with Gaussian process priors
• Bayesian inference with rescaled Gaussian process priors
• Rates of contraction with respect to \(L_2\)-distance for Bayesian nonparametric regression
• Rates of contraction for posterior distributions in \(L^{r}\)-metrics, \(1 \leq r \leq \infty\)

Cites work

• scientific article; zbMATH DE number 3126880 (Why is no real title available?)
• scientific article; zbMATH DE number 3814037 (Why is no real title available?)
• scientific article; zbMATH DE number 3824228 (Why is no real title available?)
• scientific article; zbMATH DE number 51414 (Why is no real title available?)
• A class of Wasserstein metrics for probability distributions
• Adaptive Bayesian estimation using a Gaussian random field with inverse gamma bandwidth
• Anisotropic function estimation using multi-bandwidth Gaussian processes
• Approximation, metric entropy and small ball estimates for Gaussian measures
• Asymptotic equivalence theory for nonparametric regression with random design
• Bayesian inference with rescaled Gaussian process priors
• Bernstein-von Mises theorems for Gaussian regression with increasing number of regressors
• Convergence rates of posterior distributions.
• Gaussian processes for machine learning.
• Inequalities for modified Bessel functions and their integrals
• Information Consistency of Nonparametric Gaussian Process Methods
• Information rates of nonparametric Gaussian process methods
• Lower bounds for posterior rates with Gaussian process priors
• Metric entropy and the small ball problem for Gaussian measures
• Model selection for Gaussian regression with random design
• On a Formula for the L2 Wasserstein Metric between Measures on Euclidean and Hilbert Spaces
• On posterior consistency in nonparametric regression problems
• Optimal Bayesian estimation in random covariate design with a rescaled Gaussian process prior
• Positive Definite Matrices
• Posterior consistency of Gaussian process prior for nonparametric binary regression
• Probability. Theory and examples.
• Rates of contraction of posterior distributions based on Gaussian process priors
• Small ball probabilities for Gaussian processes with stationary increments under Hölder norms
• User-friendly tail bounds for sums of random matrices

Cited in

(2)

• Projected Wasserstein Gradient Descent for High-Dimensional Bayesian Inference
• Rates of contraction of posterior distributions based on Gaussian process priors

This page was built for publication: Posterior contraction in Gaussian process regression using Wasserstein approximations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4603714)