Title :
Hierarchical overlapped SOM´s for pattern classification
Author_Institution :
Dept. of Comput. Sci. & Electr. Eng., Queensland Univ., St. Lucia, Qld., Australia
fDate :
1/1/1999 12:00:00 AM
Abstract :
We develop a multilayer overlapped self-organizing maps (SOM´s) with limited structure adaptation capabilities, and associated learning scheme for labeled pattern classification applications. The learning algorithm consists of the standard unsupervised SOM learning of synaptic weights as well as the supervised vector quantization learning. As higher layer SOMs overlap, the final classification is made by fusing the classifications of top-level overlapped SOMs. We obtained the best results ever reported for any SOM-based numerals classification system
Keywords :
handwritten character recognition; self-organising feature maps; unsupervised learning; vector quantisation; associated learning; handwritten character recognition; hierarchical SOM; multilayer overlapped self-organizing maps; numerals classification; pattern classification; structure adaptation; synaptic weights; unsupervised learning; vector quantization learning; Backpropagation algorithms; Character recognition; Merging; Multi-layer neural network; Neural networks; Neurons; Pattern classification; Self organizing feature maps; Training data; Vector quantization;
Journal_Title :
Neural Networks, IEEE Transactions on
