Activity Number:
|
213
- Lead with Statistics in Uncertainty Quantification
|
Type:
|
Invited
|
Date/Time:
|
Monday, July 30, 2018 : 2:00 PM to 3:50 PM
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
Abstract #326644
|
Presentation
|
Title:
|
Screening for Important Factors in Computer Experiments
|
Author(s):
|
David Steinberg* and Natalie Abel
|
Companies:
|
Tel Aviv University and Tel Aviv University
|
Keywords:
|
Computer Experiments;
Factor Screening;
Gaussian Process Models;
Latin Hypercube Designs;
Definitive Screening Designs
|
Abstract:
|
Factor screening is often an important first step in exploring the relationships between the inputs to a computer simulator and its outputs. Although much research has been devoted to methods for designing computer experiments and for analyzing the outcome, not much work has focused on screening the inputs. Here we carry out a largely empirical study that compares several different design and analysis strategies when the goal is initial screening. The designs include both methods proposed specifically for computer experiments like Latin hypercube designs and maximum entropy designs and methods proposed for physical experiments like the definitive screening design. On the analysis side, we compare predictive measures based on Gaussian process models and on splines and simple polynomial regression.
|
Authors who are presenting talks have a * after their name.