Screening DOE for High-Dimensionality M&S Studies

Computational modeling and simulation (M&S) enable military analysts and engineers to study complex systems as an alternative to physical experimentation which can be expensive, resource-intensive, and in some situations, impossible. However, high-dimensionality computational simulations present challenges in operations research and system analysis studies, especially when simulation run-times are high. Efficient dimension reduction allows analysts to prioritize key variables earlier on in order to better allocate computational resources toward these variables for further analysis and optimization.

This research explores the variable-screening performance of existing experimental design algorithms including space filling designs, with limited sample sizes using a variety of known test functions with high dimensionality and varying level s of complexity, both with and without noise. The author proposes a novel space filling design which improves variable selection precision in high-dimensional run-constrained simulations with noise. Preliminary results, implications, tradeoffs, and limitations are discussed, and can be immediately applied in practice to improve M&S outcomes, as well as potential extensions to active-learning applications such as AI/ML.