Hybrid algorithms for generating optimal designs for discriminating multiple nonlinear models under various error distributional assumptions.

摘要: Finding a model-based optimal design that can optimally discriminate among class of plausible models is difficult task because the criterion non-differentiable and requires 2 or more layers nested optimization. We propose hybrid algorithms based on particle swarm optimization (PSO) to solve such problems, including cases when singular, mean response some are not fully specified problems involve 4 Using several classical examples, we show proposed PSO-based criteria specific, with few repeated runs, produce either an highly efficient design. They also generally faster than current algorithms, which slow work for only specific discriminating criteria. As application, apply our techniques find designs dose-response study in toxicology 5 possible compare their performances traditional recently algorithm. In supplementary material, provide R package generate different types evaluate efficiencies competing so user implement informed