A Novel Diversity Measure of Genetic Programming

The Genetic Programming (GP)[1] is one of the novel evolutionary computation which is applied to complicated structure problems. The most important feature of GP is that individuals in GP are represented by the tree structure. The expression ability of tree structure is very flexible, and this is why GP is expected to solve many difficulties of Genetic Algorithms (GAs). However, the analysis of GP dynamics is difficult because of tree structure[2]. To solve this problem, I propose a novel diversity measure in genetic programming by means of subtree entropy. I also propose a tree simplification method which removes redundancy parts from an individual genotype. To show an advantage of our methods, the computational experiments are carried out taking a symbolic regression problem as an example.