Gibbs Reference Prior for Robust Gaussian Process Emulation
Joseph Muré · arXiv (Cornell University) · 2017
We propose an objective posterior distribution on correlation kernel parameters for Simple Kriging models in the spirit of reference posteriors. Because it is proper and defined through its conditional densities, it lends itself well to Gibbs sampling, thus making the full-Bayesian procedure tractable. Numerical examples show it has near-optimal frequentist performance in terms of prediction interval coverage.