Optimization using structured neural networks

Modeling of a system from observed data is often only the preliminary stage of a more extensive problem. An appropriate choice of model structure can greatly facilitate the solution to the subsequent problem of interest. The optimization of a static system is a specific example considered in the paper where an approximation to an unknown function has to be obtained in order to determine a point of minimum. Models (approximating functions) incorporating several neural networks are investigated where the structure of the models are chosen to give the solution to the optimization problem directly as a by-product of the modeling stage.