Projection Array--Based Designs for Computer Experiments

• Activity Number: 103
• Type: Invited
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Jason Loeppky*+ and Leslie Moore and Brian Williams
• Companies: The University of British Columbia and Los Alamos National Laboratory and Los Alamos National Laboratory
• Keywords: Computer Model ; Gaussian Process ; Latin Hypercube ; Orthogonal Array

Abstract

Computer models are extensively used to simulate complex physical processes. Deterministic computer models often are slow to run so obtaining simulation runs as a basis for construction of a statistical emulator in place of the code is appealing. In this talk a new class of designs for computer experiments are proposed. A projection array based design defines sets of simulation runs with properties that extend the conceptual properties of orthogonal array based Latin hypercube sampling, particularly to underlying design structures other than orthogonal arrays. Also, an approach to adding sequential sets of simulation runs to improve emulator performance is described that maintains dense fill of marginal projections of the design. We also discuss how criterion, such as maximin distance or integrated mean squared error can be easily incorporated into the construction of these designs.