Automatic feature selection using FS-NEAT

摘要: This article describes a series of experiments used to analyze the FS-NEAT method on double pole-balancing domain. The is compared with regular NEAT discern its strengths and weaknesses. Both find policy, implemented in neural network, solve task by use genetic algorithms. FS-NEAT, contrary NEAT, uses different starting population. Whereas networks start out links between all inputs output, have only one link an input output. It believed that this more simple topology allows for effective feature (input)-selection.