高維度不規則區域最佳均勻實驗設計的快速算法

Abstract

在實驗設計中，時常會觀察到有不規則形狀的實驗區域。在以 Central Composite Discrepancy 為實驗均勻性的衡量指標之下，我們提 出 Discrete Particle Swarm Optimization (DPSO) 最佳化演算法在一般性的實驗區域上找尋最佳化實驗設計。實驗數據顯示此演算法能比現有文獻中的最佳化演算法更有效率地找尋最佳實驗設計。為處理高維度的 Central Composite Discrepancy 龐大計算量，我們利用圖形處理器做運算上的加速而能夠在合理的時間內尋找的不錯的均勻實驗設計。

In experiment designs, irregular shapes of experimental regions are often observed. Using a recently proposed discrepancy measurement Central Composite Discrepancy as uniformity criterion, we propose a Discrete Particle Swarm Optimization algorithm for optimizing experimental designs on the general input domains. Numerical results show evidences that the new proposed algorithm is superior to other optimization algorithm in established literature. For the high computation cost of computing Central Composite Discrepancy on higher dimensions, using Graphic Processing Unit for acceleration enable us to find uniform design on higher dimensions in reasonable time.