|
Abstract:
|
Nature-inspired metaheuristic algorithms are powerful and flexible general purpose tools for solving optimizing problems. Particle swarm optimization, Competitive swarm optimizer, Differential evolutionary and Cuckoo algorithms are some examples that are already widely used in engineering and computer science but not so in mainstream statistics. This talk discusses utilities and advantages of some of these algorithms for finding optimal designs in high dimensional problems and compares their performances with current algorithms for finding optimal designs in statistics. As an application, we use Cuckoo algorithm or one its enhancements to find several types of optimal designs for the versatile five-parameter logistic model that can model asymmetries in the data. Our computed designs can be continuous or exact, and include multiple-objective optimal designs and optimal designs when there are multiple interacting independent variables in the model. When applicable, we also provide theoretical justifications for our generated designs
|
