A Sequential Information-Theoretic Approach to Design of Computer Experiments
Ali Farhang-Mehr, Shapour Azarm · 9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimization · 2002
This paper presents a new sequential informationtheoretic approach to the design of deterministic computer experiments. The approach can be used to sequentially construct a global approximation to the response from computationally expensive simulation codes. The main idea is to use the information obtained from all prior experiments to identify irregularities in the response function behavior and design new experiments accordingly. A new algorithm, referred to as Sequential MAXimum Entropy Design (SMAXED), is developed based on this approach. Details of this algorithm are presented and its capabilities are demonstrated with a simple numerical example.