|
Abstract:
|
In many engineering applications, one is interested in identifying the values of the inputs in computer experiments that lead to a response above a prespecified threshold, or in a region of interest. In this talk, we will introduce statistical methodology that identifies the desired contours (i.e., iso-surfaces) in the input space. First, a stochastic model is used to approximate the global response surface. Then, for expensive black box functions, we will propose a sequential approach for the selection of trials directed toward improvement of the estimation of the contours. Finally, extraction of the contours from the estimated surface will be outlined. Application also includes multiple contour estimation and highest posterior density region.
|
