The relationship between learning and evolution in static and dynamic environments

Abstract

Evolution and lifetime learning have been adopted by living creatures to get the best of the adaptation processes to natural environments. Within the Machine Learning (ML) arena such methods have been treated, particularly in the fields of Genetic and Evolutionary Computation and Artificial Neural Networks. Why not to combine both techniques, giving rise to several ML models, namely those based on Lamarckian or Baldwinian approaches? The results so far obtained point to better performances with the former ones under static settings, but reward the latter under dynamic environments, where the learning tasks change over time.