A novel Self-Organizing Map (SOM) learning algorithm with nearest and farthest neurons

The Self-Organizing Map (SOM) has applications like dimension reduction, data clustering, image analysis, and many others. In conventional SOM, the weights of the winner and its neighboring neurons are updated regardless of their distance from the input vector. In the proposed SOM, the farthest and nearest neurons from among the 1-neighborhood of the winner neuron, and also the winning frequency of each neuron are found out and taken into account while updating the weight. This new SOM is applied to various input data sets and the learning performance is evaluated using three standard measurements. It is confirmed that modified SOM obtained a far better result and better effective mapping as compared to the conventional SOM, which reflects the input data distribution. (C) 2014 Production and hosting by Elsevier B.V. on behalf of Faculty of Engineering, Alexandria University.