ADAPTIVE STRATEGY FOR GA INSPIRED BY BIOLOGICAL EVOLUTION

Gene duplication theory was first proposed by a Japanese biologist, Ohno, in the 1970s. Inspired by the theory, we developed a gene-duplicating genetic algorithm (GDGA) with several variants. Individuals with various gene lengths are evolved based on a parameter-free genetic algorithm (i.e., a new GA without genetic parameters set in advance) and then genes with different lengths are concatenated by migrating among sub-populations (i.e., a population was divided in advance into several sub-populations according to each gene-duplication type). To verify the algorithm's performance, we previously performed a comparative study and found a relationship between the features of the test functions and the adequate types of gene-duplication. In this study, we further describe how we have extended the scheme by automatically adapting its search strategy to various test functions without a priori knowledge of them, and verify the performance of the adaptive strategy compared to that of an adequate type of gene-duplication.