Composite grid designs for adaptive computer experiments with fast inference

• 成果类型: Article
• 署名作者: Plumlee, M.; Erickson, C. B.; Ankenman, B. E.; Lawrence, E.
• 署名单位: Northwestern University; United States Department of Energy (DOE); Los Alamos National Laboratory
• 刊物名称: BIOMETRIKA
• ISSN/ISSBN: 0006-3444
• DOI: 10.1093/biomet/asaa084
• 发表日期: 2021
• 页码: 749755
• 关键词: partial-differential-equations; stochastic collocation method; variable selection; prediction; models

摘要

Experiments are often used to produce emulators of deterministic computer code. This article introduces composite grid experimental designs and a sequential method for building the designs for accurate emulation. Computational methods are developed that enable fast and exact Gaussian process inference even with large sample sizes. We demonstrate that the proposed approach can produce emulators that are orders of magnitude more accurate than current approximations at a comparable computational cost.