Significance measure of Local Cluster Neural Networks

Abstract

Artificial neural networks are intended to be used in emerging technologies as information processing systems because their biological equivalents seem to be tolerant to internal failures of computational elements. In this paper, we introduce a measurement which can identify significant neurons of the local cluster neural network and can be used to increase the fault tolerance of this network architecture. Furthermore, it show that this technique can control the network's complexity. Moreover, by this quality different parameter sets of the network and training techniques can be judged with respect to their fault tolerant properties.