Abstract
This contribution considers some convergence and optimality properties of the Cascade-Correlation Algorithm (CCA). It is proved, that arbitrary, non-contradicting learning tasks can be solved with linear output neurons within a finite number of steps. Furthermore, it is shown that the correlation criterion proposed by Fahlman [3] does not necessarily choose optimal weights. An optimal criterion is given for linear output neurons. For nonlinear output neurons it is demonstrated, that the CCA does not need to converge even for finite learning tasks. Thus, it is generally not an universal approximation tool.
This work was supported by the Thuringian Ministry for Science, Research and Arts (project ITHERA, B511-95004).
Preview
Unable to display preview. Download preview PDF.
References
A. Doering. Optimization of Feature Extraction and Classifier Structure for Pattern Recognition with Neural Networks. PhD thesis, TU Dresden, 1997.
G.P. Drago and S. Ridella. Convergence properties of cascade correlation in function approximation. Neural Comput. and Applic., 2:142–7, 1994.
S. E. Fahlman and C. Lebiere. The cascade-correlation learning architecture. In D. S. Touretzky, editor, Advances in Neural Information Processing Systems, volume 2, pages 524–532, Denver 1989, 1990. Morgan Kaufmann, San Mateo.
J.N. Hwang, S.S. You, S.R. Lay, and I.C. Jou. The cascade-correlation learning: A projection pursuit learning perspective. IEEE Transactions on Neural Networks, 7(2):278–89, March 1996.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1997 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Doering, A., Galicki, M., Witte, H. (1997). Admissibility and optimality of the cascade-correlation algorithm. In: Gerstner, W., Germond, A., Hasler, M., Nicoud, JD. (eds) Artificial Neural Networks — ICANN'97. ICANN 1997. Lecture Notes in Computer Science, vol 1327. Springer, Berlin, Heidelberg. https://doi.org/10.1007/BFb0020205
Download citation
DOI: https://doi.org/10.1007/BFb0020205
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-63631-1
Online ISBN: 978-3-540-69620-9
eBook Packages: Springer Book Archive