Explicit class structure with closeness and similarity between neurons

We have so far introduced the concept of individually and collectively treated neurons to produce explicit class structure in SOM. Though it has produced explicit class boundaries in many well-known benchmark data, the introduction of the individually treated neurons have naturally reduced the topographical preservation. To overcome this shortcoming, we introduce closeness and similarity between neurons in learning. Neurons are more collectively connected when neurons are close and similar to each other. We applied the method to the well-known Iris and voting data in machine learning database to examine whether the new method is effective in producing explicit class structure with good topological preservation. Preliminary experimental results confirmed that class boundaries were made explicit by the interaction of ITN with CTN with closeness and similarity between neurons. In addition, improved performance could be obtained in terms of quantization, topological, training and generalization errors.