A Constructive Algorithm for Real Valued Multi-category Classification Problems
In this paper, an overview of a new constructive algorithm is proposed. Most common binary-unit based constructive algorithms are faced with 3 major drawbacks: binary inputs, only one output and a complex connectivity. The proposed algorithm aims to overcome these problems. An extension to multiple outputs of the sequential learning algorithm in combination with the Barycentric correction procedure is used. The performance of this new algorithm is evaluated in comparison with cascade correlation, on the Vowel classification benchmark.
KeywordsHide Layer Output Neuron Sequential Learning Target Vector Constructive Algorithm
Unable to display preview. Download preview PDF.
- A. Bermak and H. Poulard. On VLSI implementation of multiple output sequential learning networks. In this volume, pages 93–97.Google Scholar
- S. Fahlman and C. Lebiere. The cascade-correlation learning architecture, pages 524–532. Morgan Kaufmann, San Mateo, CA, 1990. (D. Touretzky, ed.).Google Scholar
- R. Parekh, J. Yang, and V. Honavar. Constructive neural network learning algorithms for multi-category pattern classification. Technical Report ISU-CS-TR 95-15a, Department of Computer Science, Iowa State University, 1995.Google Scholar
- H. Poulard and N. Hernandez. Two efficient constructive algorithms. Working paper, 1996.Google Scholar
- H. Poulard and S. Labrèche. A new algorithm for learning threshold unit. Technical report, LAAS-CNRS, 1996. Submitted paper, available at URL http://www.laas.fr/~poulard/papers/bcp.ps.gz.Google Scholar