Annie Yao
Title: Quasi-Oracle Arrays
Date: April 17th, 2026
Time: 10:30am
Location: LIB 7200
Supervised by: Boxin Tang
Abstract:
Computer simulations are essential for modeling complex systems, but their high computational cost requires the use of efficient surrogate models. The quality of these surrogates depends on selected input points through space-filling designs, which spread points uniformly to capture the full input space. Oracle arrays are a special class of these designs that achieve the theoretical maximum distance between all points, but they only exist for certain sets of parameters.
This thesis introduces quasi-oracle arrays, which use computational methods to approximate the optimal properties of oracle arrays when exact constructions are not possible. We evaluate these designs using the maximin Hamming distance criterion. Our research implements a design generation process involving a greedy construction algorithm, local neighborhood search, and simulated annealing. Computational results demonstrate that this approach consistently produces high-quality designs that outperform random generation.