Application of neural network to interactive physical programming

Abstract

A neural network based interactive physical programming approach is proposed in this paper. The approximate model of Pareto surface at a given Pareto design is developed based on neural networks, and a map from Pareto designs to their corresponding evaluation values is built. Genetic algorithms is used to find the Pareto design that best satisfies the designer's local preferences. An example is given to illustrate the proposed method. © Springer-Verlag Berlin Heidelberg 2005.