Q-learning chaos controller

We demonstrate the perspectives of neural networks for the challenging problem of chaos control. A self-learning neural network based controller is presented suitable for chaos control in the nonlinear control regime. Besides its intrinsic noise tolerance the main advantages of the controller consists in its ability to find the control strategy for a “black-box” system. For the purpose of learning optimal series of small control actions a Q-learning algorithm is successfully applied. In turn, our investigations suggest that chaotic systems are very well suited as test beds of reinforcement learning algorithms