Faster evolution and evolvability control of genetic algorithms using a softmax mutation method
Y. Sasaki, Hugo de GARIS · 2004
We introduce a new mutation method in evolutionary algorithms called Softmax Mutation, based on a Gibbs or Boltzmann probability distribution. Comparative experimental runs with a traditional genetic algorithm showed it to be a better alternative to the standard blind genetic operator of random mutation. The advantages of this method are not restricted to its faster evolution (namely a three fold speed up). It also impacts positively on evolvability.