Title: On algorithms for generating multiple optimal experimental designs
Authors: Yongtao Cao - Indiana University of Pennsylvania (United States) [presenting]
Abstract: The development of high performance algorithms for generating multiple optimal experimental designs is discussed. A new Pareto-based coordinate exchange algorithm for populating or approximating the true Pareto front for multi-criteria optimal experimental design problems will be emphasized. This heuristic combines an elitist-like operator inspired by evolutionary multi-objective optimization algorithms with a coordinate exchange operator that is commonly used to construct optimal designs. Benchmarking results from two to four dimensional and from screening design to split-plot design examples demonstrate that the proposed hybrid algorithm can generate highly reliable Pareto fronts with less computational effort than existing procedures in the statistics literature. The proposed algorithm also utilizes a multi-start operator, which makes it readily parallelizable for high performance computing infrastructures.