A Study on Multimodal Genetic Programming Introducing Program Simplification

In this research, we introduce a program simplification method into Multimodal Genetic Programming (MMGP) and investigate its effectiveness on multimodal program optimization benchmark. In recent years, multimodal optimization that simultaneously acquires a global and multiple local optimum solutions is studied in evolutionary algorithms (EAs). MMGP we proposed is an extension of genetic programming to multimodal optimization that simultaneously acquires a global and local optimum programs in a single run. However, since MMGP divides the solution set to several clusters depending on a tree similarity measurement, some programs cannot be assigned to appropriate cluster when a redundant subtree is generated in the optimization process. To overcome this problem, this research introduces a simplification method of a program into MMGP to remove redundant subtrees and appropriately calculate similarity of programs. The experiment that compares MMGP with and without the simplification method is conducted. The experimental result reveals that the simplification does not significantly improve the search ability of MMGP because the simplification does not much affect the optimization process of MMGP on the benchmark problem used in this research.