The T-SOM (Tree-SOM)
I introduce the T-SOM, an unsupervised neural network model based on well-known Kohonen Self-Organizing Maps. This model adds to SOM-properties the next new characteristics : a multiresolution knowledge representation, a low complexity algorithm and a simplified learning parameters tuning. A T-SOM network is a data analysis tool specially efficient in large volume data processing. The real purpose of this article is not to present one more neural network model but to show all advantages of such a hierarchical structure, both in learning and results exploitation.
Unable to display preview. Download preview PDF.
- 1.Kohonen, T.: Self-organizing Maps. Springer Eds (1995) 143–173Google Scholar
- 2.Fritzke, B.: Growing cell structures: a self organizing net for unsupervised and supervised learning. Int. Comp. Sc.Inst. California, Berkeley. (1993)Google Scholar
- 6.Zhao, Z.: Weight distance display of Kohonen maps. NeuroNîmes. (1992) 611–620Google Scholar
- 7.Lampinen, J., Oja, E.: Distortion tolerant pattern recognition based on self organizing feature extraction. IEEE Trans. on Neural Networks. 6 (1995).Google Scholar
- 8.Smagt, P.V.D., Krose, B.: Using many particle decomposition to get a parallel self-organizing map. Joint Nat. Conf. in Computer Science (1995)Google Scholar
- 9.Zbrehen, S.,Blayo, F.: A geometric organization measure for Kohonen's maps. NeuroNîmes (1992) 603–610Google Scholar
- 10.Li, K.P.: A Learning algoritm with multiple criteria for self-organizing feature map. Artificial Neural Networks (1991) 1353–1356.Google Scholar
- 11.Koikkalainen, P., Oja, E.: Self organizing hierarchical feature maps. Int. Join. Conf. on Neural Networks II 279–284.Google Scholar
- 12.Bauer, H.U., Villman, Th.: Growing a hypercubical output space in a selforganizing feature map. Int. Comp. Sc.Inst. California, Berkeley. (1995)Google Scholar