What types of problem is NEAT applicable to?
There is no definitive answer here as research into NEAT is ongoing and there have been varying degrees of success reported. Significant problem domains that I know of for which good or near-perfect solutions have been found are; Single and Double Pole Balancing, Finless Rocket Control, Prey Capture, The 3/6 and 11-Multiplexer logic problems and Pattern classification. Note however that a significant number of problems do not have a ‘perfect’ solution as such, just good and bad solutions. E.g. in game playing experiments it is common to use co-evolution where fitness is evaluated by networks competing against each other, the fitness scores are therefore only indications of relative fitness not overall ability to play a game. One very basic requirement when considering an experiment is that a fast method of evaluating fitness must be available for the problem at hand. This is so that NEAT can evaluate as many candidate networks as possible within a given time period. The more evaluation
