| Summary: | 碩士 === 國立高雄大學 === 統計學研究所 === 100 === We consider a contour estimation problem, in which the responses are evaluated
by costly iterative processes. A two-stage approach is proposed here to estimate contour lines e ciently. In the rst stage, a surrogate-assistant approach used to quickly estimate the possible target contours. To save time on function evaluations, we propose a dynamic control approach to compute the function evaluation iterations by adaptively determining whether each of the function evaluation iterations should be paused or kept running. After identifying possible contours from the surrogates constructed by a Gaussian process type method, in the second stage, we introduce a window tracking approach to trace out the true contours with higher accuracy. Several numerical experiments and a real example in a queueing system are used to illustrate the performance of the new approaches.
|
|---|
