Multipopulation cooperative coevolutionary programming (MCCP) to enhance design innovation

This paper describes the development of an evolutionary algorithm called Multipopulation Cooperative Coevolutionary Programming (MCCP) that extends Genetic Programming (GP) to search for a set of maximally different solutions for program induction problems. The GP search is structured to generate a set of alternatives that are similar in design performance, but are dissimilar from each other in the solution (or design parameter) space. This is expected to yield potentially more creative designs, thus enhancing design innovation. Application of MCCP is demonstrated through an illustrative example involving GP-based classification of genetic data to diagnose malignancy in cancer. Four different classifiers, based on highly dissimilar combinations of genes, but with similar prediction performances were generated. As these classifiers use a diverse set of genes, they are collectively more effective in screening cancer samples that may not all properly express every gene. Categories a...

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Emily M. Zechman, S. Ranji Ranjithan
• Keywords: Cooperative Coevolutionary Programming, GECCO 2005, Genetic Programming, Keywords Genetic Programming