A new model to realize variable size Genetic Network Programming
Hironobu Katagiri, Kotaro Hirasawa, Jinglu Hu, Junichi Murata · 2002
Genetic Network Programming (GNP) [1] is an exten-sion of Genetic Programming motivated by the strong expression ability of graph. A program in GNP is an arbitrary directed graph, composed of nodes connected to each other by directed arcs. Figure 1 shows the basic scheme of GNP system. Previously, the program size of GNP was xed. In the paper, a new method is pro-posed, where the program size is adaptively changed depending on the frequency of use of nodes. Generally, large programs have high expression ability, while their evolutions are disturbed by their enormous search s-paces, also they occupy many memory resources and consume much calculation time. To control and to de-cide a proper program size are important and diÆcult problems in Evolutionary Computation. We introduce