Local response neural networks and fuzzy logic for control

It is shown how to build and train multilayer perceptrons for the approximation of control functions. The special class of perceptrons called local response networks have the advantage that they train much faster than the general multilayer perceptrons (MLPs), and that the accuracy of the approximation can be increased by adding more neurons without the need of global retraining. They also have the advantage that the knowledge of a trained network is easily translated into rules, similar to fuzzy logic.<>