SOAN: Self organizing with adaptive neighborhood neural network
In this work we describe the design and functioning of a new neural network based on vector quantification. This network, which we call SOAN (Self Organizing with Adaptative Neigborhood) has a greater degree of learning flexibility due to the use of an interaction radius between neurones which varies spatially and temporally, and an adaptative neighbourhood function. Secondly, we have introduced mechanisms into the network with the aim of guaranteeing that all of its neurones contribute as far as possible in reducing the quantification error. Finally, we have carried out several experiments obtaining highly favourable results, which after having been contrasted with those obtained with the application of the SOM network, confirm the utility and advantages of our approach.
Unable to display preview. Download preview PDF.
- 1.T. Kohonen. Self-Organizing Maps. Springer Series in Information Sciences, Vol. 30. Springer-Verlag.Google Scholar
- 2.M. Afzal Upal, E. M. Neufeld, Comparison of Unsupervised Classifiers. In Proceedings of ISIS’96:342–353, World Scientific.Google Scholar
- 3.J. Muñoz Aprendizaje competitivo para la cuantificación vectorial. In Proceedings of CAEPIA’97:377–385, 1997. In spanish.Google Scholar
- 4.T. Kohonen, J. Hynninen, Jari Kangas and J. Laaksonen. SOM-PAK: The Self-Organizing Map Program Package. Laboratory of Computer and Information Science, Faculty of Information Technology, Helsinki University of Technology, Report A31, Otaniemi 1996.Google Scholar
- 5.Keinosuke Fukinaga. Introduction to Statistical Pattern Recognition. Computer Science and Scientific Computing, W. Rheinboldt and D. Siewiorek (Eds.) Academic Press.Google Scholar
- 6.R. Schalkoff Pattern Recognition: Statistical, Structural and Neural Approaches. John Wiley & Sons.Google Scholar
- 7.Gail A: Carpenter y Stephen Grossberg. ART 2: Self-organization of stable category recognition codes for analog input patterns. Applied optics, 26(23):4919–4930.Google Scholar