Abstract
Multilayer perceptrons are learning structures often used in the connectionist approach. A backpropagation algorithm, which enables learning, is a fixed point research algorithm. As such, it induces the various behaviours of chaotic dynamics. One can apply to it the tools of chaos theory. Amongst the useable tools, a measurement of behaviour, or more precisely stability, exists, namely Lyapunov numbers. The obtaining of these numbers comes through awareness of the eigen values of the jacobian matrix associated to the weight modification functions. We give a method of calculation whose efficiency comes from the use of the particularities of the backpropagation. From the calculation of the Lyapunov numbers, the basic backpropagation algorithm is modified. We propose the first part of a new learning algorithm whose originality resides in a strategy of gradient step constraint, arising from the obligation of a stable behaviour. Its values is related to obtaining rapid convergence.
This is a preview of subscription content, log in via an institution to check access.
Preview
Unable to display preview. Download preview PDF.
Similar content being viewed by others
References
M. Milgram-Reconnaissance des formes, méthodes numériques et connexionnistes-Armand Colin, 1993.
Y. Le Cun-Modèles connexionnistes de l'apprentissage-Thèse de doctorat, Université de Paris VI, 1987.
B. Widrow, & al.-30 years of adaptative neural networks: perceptron, madaline and backpropagation-Procedings of the IEE, vol. 78, n∘ 9, p. 1415–1442, IEE, 1990.
S. Shah, F. Palmieri, M. Datum-Optimal filtering algorithms for fast learning in feedforward neural networks-Neural Networks, vol. 5, p. 779–787, Pergamon Press, 1992.
R.A. Jacobs-Increased rates of convergence through learning rate adaptation — Neural Networks, vol. 1, p. 295–308, Pergamon Press, 1988.
Y. Lee, S.H. Oh, M.W. Kim-An analysis of premature saturation in backpropagation learning-Neural Networks, vol. 6, p. 719–728, Pergamon Press, 1993.
P. Lascaux, R. Théodor-Analyse numérique matricielle appliquée à l'art de l'ingénieur-Masson, 1986.
C. Brezinski-Algorithmique numérique-Ellipses, 1988.
P. Frederickson, J.L. Kaplan, E.D. Yorke, J.A. Yorke-The Lyapunov dimension of strange attractors-Journal of Differential Equations, vol. 9, p. 185–207, Academic Press, 1983.
A. Grorud, D. Talay-Approximation of Lyapunov exponents of non-linear stochastic differential systems-Rapport de recherche n∘ 1341, INRIA, 1990.
J. Weitkämper-A study of bifurcations in a circular real cellular automaton-Department de Matematica aplicada i analisi, Universitat de Barcelona, 1991.
F. Zou, J.A. Nossek-Bifurcation and chaos in cellular neural networks-IEEE transactions on Circuits and Systems: Fondamental Theory and Applications, volume 40, n∘ 5, pages 166–173, IEEE, 1993.
D. Guegan-Notion de chaos, approche dynamique et problèmes d'identification-Rapport de recherche n∘ 1623, INRIA, 1992.
P. Maneville-Structures dissipatives, chaos et turbulence-Collection Aléa Saclay, 1991.
B. Karayiannis, A.N. Venetsanopoulos-Efficient Learning Algorithms for Neural Networks-IEEE Transactions on systems, man and cybernetics, vol. 23, n∘ 5, p. 1372–1383, 1993.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1995 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Augereau, B., Simon, T., Bernard, J., Heit, B. (1995). The BP-λL1 algorithm: Non-chaotic and accelerated learning in a MLP network. In: Mira, J., Sandoval, F. (eds) From Natural to Artificial Neural Computation. IWANN 1995. Lecture Notes in Computer Science, vol 930. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-59497-3_180
Download citation
DOI: https://doi.org/10.1007/3-540-59497-3_180
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-59497-0
Online ISBN: 978-3-540-49288-7
eBook Packages: Springer Book Archive