Technometrics Highlights
May Issue Has Improved Web Format
when they prepare manuscripts computer simulator. Uncertainty
for submission to Technometrics. regarding the values of some of the
They also should refer to the sup- simulator inputs generates uncer-
plemental material in the text of tainty about the model outputs;
their manuscripts and include the most of the research in this area
relevant files as part of the package highlights the effect each input
that is uploaded for review. Details has on the model output variance.
about how to prepare the supple- Oakley emphasizes, instead, the
mental materials can be found at effect of uncertainty on the result
http://pubs.amstat.org/page/tech/ of the decision process, aided by
information-for-authors. decision-theoretic measures he
The May 2009 issue of develops. Computation of these
Technometrics leads with two measures can be impractical for
articles that will be presented complex computer models, and
in the Technometrics session at so efficient computational tools
JSM this summer. The first is using Gaussian processes also are
F
or many of our rfdeaders, by Brian J. Reich, Curtis B. presented. An illustration is given
Technometrics’ web site Storlie, and Howard D. Bondell in the field of medical decision-
(http://pubs.amstat.org/loi/ and is titled “Variable Selection making, and the Gaussian process
tech) is a primary access point to in Bayesian Smoothing Spline approach is compared with con-
the journal. A recent upgrade of ANOVA Models: Application to ventional Monte Carlo sampling.
the ASA site has led to an Deterministic Computer Codes.” Computer experiments are also
improved electronic version of the With many predictors, choosing the subject area of our third article,
journal. The online PDF files an appropriate subset of the cova- “Adaptive Design and Analysis of
now include dynamic linking of riates is a crucial, and difficult, step Supercomputer Experiments,” by
email addresses and references in nonparametric regression. This Robert B. Gramacy and Herbert
and have bookmarks with all sec- paper proposes a Bayesian non- K. H. Lee. Computer experi-
tion and subsection titles. parametric regression model for ments are often a cost-effective
A major new addition is the curve-fitting and variable selection. method of allowing modeling of
ability to publish supplemental The smoothing spline ANOVA a response surface of a physical
material, available only on the framework is used to decompose experiment that is too costly or
web site. These files might include the regression function into inter- difficult to run in the lab or field.
software for implementing new pretable main effect and interac- Running the computer model over
statistical methods, detailed proofs tion functions. Stochastic search a dense input grid can be prohibi-
or derivations, additional simula- variable selection via MCMC sam- tively expensive, yet running over
tions or tables of results, or data pling is used to search for models a sparse design chosen in advance
sets. This supplemental material is that fit the data well. Variable selec- can result in insufficient informa-
great added value for readers and tion is found to be highly sensitive tion in parts of the space, particu-
authors. In particular, providing to hyperparameter choice, and larly when the surface calls for a
electronically accessible code with the authors develop a technique nonstationary model. Gramacy
an article will accelerate the dis- to select hyperparameters. The and Lee propose an approach that
semination of new methods in method is used to build an emula- automatically explores the space
applied statistical work. tor for a complex computer model while simultaneously fitting the
Our first article with supple- for two-phase fluid flow. response surface, using predictive
mental material will appear in the The second paper, by Jeremy uncertainty to guide subsequent
May issue. To access supplemental Oakley, considers problems in experimental runs. They use the
files, just log in to the online ver- which a computer model is used as Bayesian treed Gaussian process
sion of Technometrics. There, you a decision support tool. The title of as the surrogate model. An adap-
will find a link to the supplemental this paper is “Decision-Theoretic tive sequential design framework
material alongside the link to the Sensitivity Analysis for Complex is developed to cope with an asyn-
PDF version of the article. Computer Models.” The basic chronous, random, agent–based
Authors should think about idea is to discover good solutions supercomputing environment
including supplemental material to problems being studied with a by using a hybrid approach that
MAY 2009 AMSTAT NEWS 23
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84