Plastic self-organizing map

Good examples of dynamic neural networks are the cascade correlation (Fahhnan and Lebiere, 1990), which extends the MLP to add structure to its network the growing neural gas with utility (GNG-U) (Fritzke, 1997), which uses a SOM-like structure to leam the shape of the data and the plastic self-organizing map (PSOM) (Lang and Warwick, 2002), which is very similar to the SOM except that the neurons are not fixed in a grid and new neurons may be added and removed. The PSOM will be described here. 

The plastic self-organizing map (PSOM) is a SOM-like structure consisting of neurons and links. Like the SOM, a neuron contains a reference vector, but unlike the SOM each neuron is not bound in a grid. Instead, each neuron is connected to other neurons using a dynamic link (Figure 4.14). 

FIGURE 4.14 The structure of the plastic self-organizing map (PSOM) showing four neurons, four links and their reference vectors. Class numbers are shown in boxes next to the neuron. 

Lang, R.I.W. and Warwick, K. (2002) The plastic self-organizing map. In WCCI Proceedings of the International Joint Conference on Neural Networks (ed D. Fogel), IEEE, Honolulu, pp. 727-732. 

When new material is added to training sets there is no need to recompute a high-dimensionality to low-dimensionality transform. In fact, like Kohonen self-organizing maps (Kohonen, 2001) and the closely related plastic self-organizing map (PSOM) (Lang, this volume), NNC effectively trains in real time. 

Prediction-set samples were then passed to the network and assigned to the nearest neuron. They are represented as superscripts in Figure 9.18. All 19 prediction-set samples were assigned to neurons that had spectra with the same class label. The validation of the self-organizing maps using these 19 prediction-set samples implies that information is contained in the Raman spectra of the plastics characteristic of sample type. Classification of the plastics by the self-organizing map was as reliable as that obtained by other methods, and more importantly, the classification was obtained without any preassumptions about the data. 

---