Structural Learning by Genetic Algorithm with Damaged Genes

T. Takahama and S. Sakai (Japan)


Structural Learning, Damaged Genes,Genetic Algorithm, Neural Network


In this paper, we propose a new method of structural learning, Genetic Algorithm with Damaged Genes (DGGA). When genes are damaged, an individual who has the damaged genes may express the phenotype of the genes imperfectly, or even may not express the phenotype. To realize this phenomenon, we give a new map ping function from genotype to phenotype, which depends on damaged genes. We also introduce the probabilistic changes from a normal gene to a damaged gene. We can reduce the genes that have lower effectiveness by these changes. Through structural learning of a polynomial model and layered neural networks, we show that DGGA can optimize the parameter structure of the models and DGGA is a general-purpose method for structural learning.

Important Links:

Go Back