Multiple Self-Organizing Neural Networks with the Reduced Input Dimension

Abstract

Listen

Translate article

In this paper, we propose a new multiple self-organizing neural network architecture. Whereas conventional learning methods utilize the full dimension of the original input patterns, the proposed system consists of multiple neural networks with the reduced input dimension. We have developed three consensus schemes so as to judge the classification using multiple neural networks. Each network of ensemble has dynamic properties. The number of output neurons is increased as learning proceeds. Every output neuron has its own class threshold, which represents a class boundary. The class threshold value is tuned according to input pattern distribution. The performance of the multiple self organizing neural network is compared with those of conventional competitive learning algorithms.