Genetic Algorithms As Hill-climbing Methods
Hiroki Yoshizawa, Shuji Hashimoto · Proceedings of the ISCIE International Symposium on Stochastic Systems Theory and its Applications · 1998
In this paper, we propose a new viewpoint of search space in Genetic Algorithms (GAs) where the population hill-climbs. This corresponds to the theory of Neural Networks, where Neural Networks are understood as hill-climbing energy function (ex. the square of error). The methods of this research are as follows. We model GAs using expectation. The propriety of using expectation is one significant arguing point. Then, we illustrate the transition and the final state of population. The time evolution of GAs can be understood as a sort of hill-climbing in this sense. Last, we show that the proposed model can explain phenomena in practical trials (ex. dependency on initial population).