Efficient Constructive Techniques for Training Switching Neural Networks

In this paper a general constructive approach for training neural networks in classification problems is presented. This approach is used to construct a particular connectionist model, named Switching Neural Network (SNN), based on the conversion of the original problem in a Boolean lattice domain. The training of an SNN can be performed through a constructive algorithm, called Switch Programming (SP), based on the solution of a proper linear programming problem. Since the execution of SP may require excessive computational time, an approximate version of it, named Approximate Switch Programming (ASP) has been developed. Simulation results obtained on the StatLog benchmark show the good quality of the SNNs trained with SP and ASP.