基於雙重精準度輔助曲面的等高線估計

Abstract

我們利用兩個不同精準度的實驗結果建立一個近似高精準度實驗的高斯過程輔助曲面，並用於估計一複雜函數的等高線。主要目標是在使用相同或更少的計算成本下，利用雙精準度的輔助曲面來改進估計的效率。在循序增加實驗點改進被估計等高線的過程中，我們提出一個標準來決定下一個實驗點要進行高精準度或低精準度的實驗。一些衡量方法被用來評估被估計等高線的品質。本方法用於估計的一個網路排程系統的平均延遲時間。

We used two experimental results with different fidelity to build surrogates which is employed to estimate the contours of a complex function. Our goal is to improve efficiency by using the same or less computational budgets. We propose a criterion to decide which level of the code to perform at added points for sequentially refining the estimated contour. Some discrepancy measures are used to assess the quality of estimated contours. The proposed method is used to estimate the average delay time in a queuing network system.