Modeling and Analysis of Genetic Algorithms Using Neural Networks

摘要

Vose's genetic algorithm model assuming an infinite population is useful for a theoretical analysis. However, it is generally difficult to know transitions of infinite populations. In this paper, we propose a method for modeling genetic algorithms for infinite populations by using neural networks. We use a neural network for estimating deterministic transitions of infinite populations from stochastic data obtained through observing a process of a genetic algorithm for finite populations. Then the trained network approximates a mapping (or a vector field) which characterizes the genetic algorithm. Our method introduces a framework for analyzing genetic algorithms from the viewpoint of neural networks. In this paper, we use a mixture-of-experts architecture for modeling and show that an optimization problem, which the genetic algorithm solves, is represented as a combination of some other optimization problems corresponding to expert networks.