How can Artificial Neural Networks help making the intractable search spaces tractable

In this paper, we propose the incorporation of artificial neural network (ANN) based supervised and unsupervised machine learning techniques into the evolutionary search, in order to detect strongly connected variables. The cost of extending a search method with an ANN based learning skill is relatively low, the memory requirements and model building cost being at most linearithmic in the number of variables. As a case study, we show how these mechanisms can enable the simple (1+1) evolutionary algorithm to efficiently solve hard problems, which are provably intractable using just fixed representation and problem independent operators. Furthermore, simulation results show, that on test suites characterized by strong variable coupling, the ANN extended (1+1) evolutionary algorithm qualitatively outperform the best known, full-featured, population based estimation of distribution algorithms.