The introduction of hidden units to optimizing neural networks

A systematic approach to deriving neural network algorithms that promotes the introduction of hidden units and defines their best features is presented. To illustrate the method, the authors derive a neural network for the NP-hard integer knapsack-packing optimization problem. Preliminary simulation results have been promising, especially in light of the ease with which viable network parameters have been found. In the course of subsequent research, the authors have found the method to be applicable to bin-packing, multiprocessor scheduling, and job-sequencing problems as well