The best way to train a neural network is with a combination of reward and punishment.