A0831
Title: Uniform projection designs via differential evolution with evaluation through prediction and function maximization
Authors: Samuel Onyambu - University of California, Los Angeles (United States) [presenting]
Hongquan Xu - University of California Los Angeles (United States)
Abstract: Uniform projection designs are valued in computer experiments for their favorable low-dimensional projection properties and robustness across a variety of performance criteria. The focus is on the construction and evaluation of such designs using the differential evolution algorithm, which is known for its effectiveness in complex optimization tasks. The performance of differential evolution is compared to simulated annealing in the construction of maximin distance designs, with results indicating that differential evolution typically yields higher quality designs. A systematic analysis of the differential evolution hyperparameter space is conducted to identify configurations that promote stability and design quality. In addition, the influence of initial design choices on sequential prediction and optimization is investigated using Gaussian process models and active learning strategies. While performance depends on the setting, uniform projection designs constructed via differential evolution often demonstrate stronger and more consistent behavior compared to traditional space-filling designs. These findings support the use of uniform projection designs as a practical and effective option in high-dimensional computer experiments.
