An improved adaptive neural network and its application on random shape

Abstract

The random shape generation method is put forward based on adaptive neural networks. The adaptive neural network is trained from an arbitrary regular geometric shape during the random deformation process. Thus, the regular shape can be changed to an irregular one with the adaptive learning method, and the global and local controllability can both be enhanced. With an improvement on the traditional adaptive neural network algorithm, certainty and randomness can be fully combined, so that fuzzy controllability and adjustability can be dominated easily and concisely.