Optimization of Neural Network Topology and Information Content Using Boltzmann Methods.

摘要

Reduction in the size and complexity of neural network applications are the driving force behind the current research in network optimization. Most of the known optimization methods heavily rely on weight sharing concepts for pattern separation and recognition. The method used in the research focuses on network topology and information content for optimization. The authors have studied the change in the network topology and its effects on information content dynamically during the optimization of the network. The changes in the network topology were achieved by altering the number of weights. The primary optimization was scaled conjugate gradient and the secondary method of optimization a Boltzmann method. The findings demonstrate that for a difficult character recognition problem, the number of weights in a fully connected network can be reduced by 90.3% with a temperature of 0.55 while achieving training and testing of identical accuracies.