A1664
Title: Computational construction of sequential efficient designs for the second order model
Authors: Norah Alshammari - student (Australia) [presenting]
Stelios Georgiou - RMIT University (Australia)
Stella Stylianou - RMIT University (Australia)
Abstract: Sequential experiment designs optimize data collection by enabling efficient decisions on whether to continue or stop testing, minimizing resource usage and accelerating goals. Sequential Latin hypercube designs (SLHDs) progressively add design points, reducing computational demands. Unlike traditional model-free LHDs, this approach generates optimal designs for specified models, focusing on the second-order model used in response surface methodology to improve A-efficiency. Challenges in designing efficient high-dimensional experiments are addressed by relaxing the condition of no-point replication in equally spaced intervals. This relaxation maintains space coverage while allowing more flexibility for model-specific efficiency. Sobol sequences are used to iteratively select points that maximize the A-criterion for the second-order model. Results show superior performance compared to methods that minimize inner-point distances, offering practical guidance for selecting optimal experimental designs.
