MAES Median Selection Strategy


The main application field of the Median Selection Strategy operator are steady state algorithms. A standard steady-state ES is equivalent to a (mu + 1) ES. Only one individual is generated and evaluated at each step and gets immediately integrated into the population. Compared to generation based algorithms the information of new evaluated individuals can be integrated directly into the optimization process. The idea is to approximate the selection mechanism of a standard (mu,lambda) ES, by using a fitness buffer containing fitness values of the last n evaluations. Given a relative rate of acceptance r=mu\lambda. A newly evaluated individual substitutes the worst individual of the population, if it has a better fitness than the r*n best individuals in the buffer.