Recent Developments in Gradient-Enhanced Kriging

• Activity Number: 611
• Type: Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Peter Marcy*+
• Companies: University of Wyoming
• Keywords: computer experiments ; gradients ; dimension reduction ; design ; derivative information ; exploratory data analysis

Abstract

In the now well-established field of computer experiments, output from complex computer models is approximated by Gaussian Process (GP) surrogates. The GP framework upon which these statistical approximations (sometimes called emulators) rely allows for the incorporation of derivative information produced by the original computer code. This supplementary information can greatly improve the emulators, while also providing a better understanding of the original computer model. In this presentation I will discuss the use of emulators incorporating first partial derivatives ("Gradient-Enhanced Kriging"). I will very briefly cover design strategies, exploratory data analysis, dimension reduction, and overall utility of gradient information.