A Simulated Elastic-Force Analysis of SOM on Topological Ordering
Xiangqian Ding · Qingdao Haiyang Daxue xuebao · 2005
It is a difficult problem to analyze the mechanism of the ability of keeping topology orders by the SOM algorithm. This paper defines two kinds of simulated elastic-forces, which are in-directly proportional to the distance between pattern vectors and weight vectors, borne by a neuron i in the output lattice of Kohonen networks. One is the pull-force of an input pattern-vector to the winning neuron i and the other is the affect-force of the winning neuron j to the neighboring neuron i. The proposed elastic-forces have nine important properties. The mechanism of the evolvement of the weight vectors and the topological ordering of Kohonen's SOM may be understandably and visually explained by the new method described in this paper.