Speaker: 李春亚, 上海工程技术大学

Inviter: 熊世峰

Title: Enhancing High-dimensional Space-Filling Designs: A Geometric Approach to Boundary Optimization in Computer Experiments

Language: Chinese

Time & Venue: 2025.09.18 20:00-21:00 腾讯会议 598-368-541

Abstract: Computer experiments have indispensable roles across science and engineering, providing an effective alternative to physical testing when experiments are prohibitively expensive, time consuming, or infeasible. As computational models become more complex and the number of input parameters grows, designing experiments that guarantee comprehensive and uniform input‐space coverage is essential for constructing accurate surrogate models. We propose a boundary enhanced space filling design methodology that systematically increases sampling density along the domain’s periphery without degrading overall uniformity. Our method combines geometric analysis to quantify central clustering severity with a heuristic spatial transformation algorithm that iteratively identifies centrally located points and repositions them toward the boundaries. This process employs a randomized step size under clipping constraints to maintain feasibility within the hypercube.