確率オートマトンの学習機能を活用した遺伝的アルゴリズムの収束性の改善

抄録

It has been reported that after prolonged starvation, bacterial cells increase in the frequency of mutation and produce new advantageous phenotypes. This result gives a suggestion for the improvement of the genetic algorithm for optimization problems. The stochastic automaton is applied for the learning of the position of loci at chromosome so that the evaluation takes a higher value. The state stochastic vector of the automaton thus obtained generates a mutation rate for corresponding locus of the chromosome. This procedure helps the algorithm to escape from the trapping in local minima. The effectiveness is shown especially for the maximum search problem of a variable-separable multi-peak function.