Adaptive Latin Hypercube Designs for Computer Experiments

• Activity Number: 627
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Quality and Productivity Section
• Author(s): Ying Hung*+
• Companies: Department of Statistics, Rutgers University
• Keywords: Latin hypercube sampling ; Rao-Blackwell method ; Sequential design

Abstract

Most of the existing designs in computer experiments are chosen in advance, which results in insufficient information. To overcome this problem, a new class of adaptive designs is introduced. It is model-free and space-filling; therefore, it is attractive for complex computer experiments particularly in the early stage where model fitting tends to be unreliable. The construction procedure and design-unbiased estimators are discussed. Moreover, a refinement of the proposed design is introduced which provides better control over sample size and avoids replicates in the final sample. An application to a simulation study of the piezoelectric energy harvester shows that the proposed adaptive designs are more efficient than the existing non-adaptive designs in reducing the estimation variance.