Title: Reduction of Model Complexity and the Treatment of Discrete Inputs in Computer Model Emulation

Speaker: Curtis Storlie, University of New Mexico

Date/Time: Monday, November 16, 2009 at 9:30 – 10:30am        

Location: CSRI Building, Room 90 (Sandia NM)

Brief Abstract: The analysis of many physical and engineering problems involves running complex computational models (computer codes).  With problems of this type, it is important to understand the relationships between the input variables (whose values are often imprecisely known) and the output as well as characterize the uncertainty in the output due to the uncertainty in the inputs.  A computational model that sufficiently represents reality is often very costly in terms of run time.  The input vector to the model can be of very high dimension and complex structure.  In addition, some of the input variables are also discrete in nature (e.g. pointer variables to alternative models or different types of material, etc.).  When the models are computationally demanding, emulator approaches to their analysis have been shown to be very useful.  However, most existing approaches to this problem do not work well with a large number of input variables and/or they do not explicitly allow for discrete input variables.  In this presentation, we propose two new approaches that tackle high dimensionality through variable selection and functional ANOVA model construction while also accounting for discrete valued inputs.

CSRI POC: James Stewart, (505) 844-8630