A “Futurist” approach to dynamic environments

The optimization of dynamic environments has proved a difficult area for Genetic Algorithms. As standard haploid populations find it difficult to track a moving target, different schemes have been described to improve on the situation. We propose a novel approach by making use of a metalearner which tries to predict the next state of the environment, i.e. the next value of the goal the individuals have to achieve, by making use of the accumulated knowledge from the past performances.