Order Formation in Learning Nonlinear Robust Control Systems by Use of Neural Networks

In designing controllers, robustness against uncertainties is very important property, therefore we proposed design methods of nonlinear robust control system by use of neural networks. Competition between a neural network, which acts as a robust controller and uncertainties of the plant, plays a very important role to be robust, and simple order formation, such as power law scaling, can be found in a set of trained robust controllers. Such simple order is widely found in nature and complex systems, but learning methods without considering uncertainties has never formed such relations. Therefore competitive learning will have important relationship to order formation in various systems.