Kriging, more descriptively referred to as Gaussian process regression, is a popular Bayesian technique for constructing a surrogate model of an expensive function with quantified uncertainty. In the basic implementation of kriging, the Bayesian evidence takes the form of a finite number of evaluations of the unknown function. In this talk, I will review cokriging methods which use two additional forms of evidence: 1. evaluations of the function's gradient and 2. evaluations of a potentially cheaper model with some unknown correlation to the function of interest. Both of these observation types have much practical utility for accelerating the construction of an accurate and precise surrogate model.

Hybrid: Math, 402/Zoom   https://arizona.zoom.us/j/83541348598   Password: BB2022