A Supervised Self-Organizing Map for Structured Data
A self organizing map (SOM) for processing of structured data, using an unsupervised learning approach, called SOM-SD, has recently been proposed. Here, we suggest a new version of SOM, using the supervised learning approach. We compare the supervised version and the unsupervised version of SOM-SD on a benchmark problem involving visual patterns. As may be expected, the supervised version is able to solve the classification problem using very compact networks.
KeywordsRoot Node Leaf Node Directed Acyclic Graph Target Class Winning Neuron
Unable to display preview. Download preview PDF.
- 1.A.M. Bianucci, A. Micheli, A. Sperduti, and A. Starita. Analysis of the internal representations developed by neural networks for structures applied to qsar studies of benzodiazepines. Journal of Chemical Information and Computer Sciences, 41(1):202-218, 2001.Google Scholar
- 4.C. Goller. A Connectionist Approach for Learning Search-Control Heuristics for Automated Deduction Systems. PhD thesis, Technical University Munich, Computer Science, 1997.Google Scholar
- 5.Markus Hagenbuchner and Ah Chung Tsoi. The traffic policeman benchmark. In Michel Verleysen, editor, European Symposium on Artificial Neural Networks, ISBN 2-9600049-9-X, pages 63-68. D-Facto, April 1999.Google Scholar
- 6.Teuvo Kohonen. Self-Organisation and Associative Memory. Springer, 3rd edition, 1990.Google Scholar