EcoSta 2023: Start Registration
View Submission - EcoSta2023
A0192
Title: Efficient Kriging using designs with low fill distance and high separation distance Authors:  Xu He - Chinese Academy of Sciences (China) [presenting]
Abstract: Kriging is a powerful technique to emulate computer experiments. Accurate Kriging requires good experimental designs. The purpose is to construct space-filling designs that are most suitable for Kriging. Firstly, the study provides evidence that although it is well-known that designs with low fill distance or high separation distance are appealing for Kriging, the two distance measures should be combined to reach a more striking criterion for selecting designs. Secondly, a method is proposed to construct optimal designs under the newly proposed criterion efficiently. Thirdly, a robust method is provided to transform design points towards boundary facets of the input space. Numerical results suggest that the new criterion, construction, and transformation combined perform robustly well in Kriging interpolation under various scenarios.