STA 294: SPECIAL TOPICS
Professor Jerome Sacks, Instructor
TTh 2:15 - 3:30pm
025 Old Chemistry Building
Topic: USE AND VALIDATION OF COMPUTER/SIMULATION MODELS
Computer models representing real phenomena are ubiquitous. Two important
questions arise: How to measure the model's faithfulness in representing
reality? How to measure uncertainty (stemming from a variety of sources
including the model's fidelity to reality) in model predictions? The
course will address these two questions.
Components of the course:
I. Motivating examples (drawn from geoscience, engineering, and
biology).
II. Approximations to computer models: Approximations are
necessary for models that cannot feasibly be run many times. The emphasis
will be on approximations that use Gaussian stochastic processes for a
"Bayesian numerical analysis". The methods and implications for
interpolation, integration and optimization will be developed in the
course.
III. Design of computer experiments (selection of inputs at which
to run a model)
IV. Sensitivities/Screening: Which inputs are important/ which
can be ignored?
V. Validation - principles and strategies
Prerequisites: Most of the methodology will be developed in the
course but substantial familiarity with "principles" and ideas of
regression will be essential. The course is open to faculty and research
staff.
For further information please contact J. Sacks before August 27, 2001
at sacks@niss.org or 919-685-9313.