Abstract
In this paper, a method for pruning hidden neurones is presented, and illustrated on two different problems. It is based on the statistical study of the derivatives of the outputs of the model with regards to each hidden neurone. We claim that if the model is not using a particular neurone to estimate its outputs, then the corresponding sensitivities will have a low degree of significance. This article is an extension of a previous work dedicated to the selection of input variables. We consider each hidden layer as the input layer of a smaller network made of all the remaining layers between this one and the output. The aim of this analysis is the selection of an appropriate subset of neurones for each layer, to finally obtain a more parsimonious model.
The work of T. Czernichow has been supported by a CIFRE grant N∘91/93 with Electricité de France, Direction des Etudes et Recherches (EDF-DER).
Preview
Unable to display preview. Download preview PDF.
References
T. Czernichow, A. Muñoz, “Variable Selection through Statistical Sensitivity Analysis: Application to feedforward and recurrent networks.,” INT-SIM, http://www-sim.int-evry.fr/People/Czernichow.html, Technical Report 95-07-01, 1995.
B. Dorizzi, G. Pellieux, F. Jacquet, T. Czernichow, A. Muñoz, “Selecting the relevant variables to forecast the French T-Bond,” presented at Third Chemical Bank/Imperial College Conference on Forecasting Financial Markets, London, 1996.
P. Cardaliaguet, G. Euvrard, “Approximation of a Function and its Derivatives with a Neural Network,” Neural Networks, vol. 5, pp. 207–220, 1992.
K. Hornik, M. Stinchcombe, H. White, “Universal Approximation of an unknown Mapping and its Derivatives Using Multilayer Feedforward Networks,” Neural Networks, vol. 3, pp. 551–560, 1990.
A. R. Gallant, H. White, “On learning the derivatives of an unknown Mapping with Multilayer Feedforward Networks,” Neural Networks, vol. 5, pp. 129–138, 1992.
D. G. Luenberger, Linear and non-linear programming, 2nd ed: Adddison-Wesley, 1984.
C. D. Liu, J. Nocedal, “On the limited memory BFGS method for large scale optimization,” Mathematical programming, vol. 45, pp. 503–528, 1989.
T. Czernichow, A. Piras, K. Imhof, P. Caire, Y. Jaccard, B. Dorizzi, A. Germond, “Short Term Electrical Load Forecasting with Artificial Neural Networks,” Int. Journal of Eng. Int. Syst., To appear, 1996.
Author information
Authors and Affiliations
Editor information
Rights and permissions
Copyright information
© 1996 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Czernichow, T. (1996). Architecture selection through statistical sensitivity analysis. In: von der Malsburg, C., von Seelen, W., Vorbrüggen, J.C., Sendhoff, B. (eds) Artificial Neural Networks — ICANN 96. ICANN 1996. Lecture Notes in Computer Science, vol 1112. Springer, Berlin, Heidelberg. https://doi.org/10.1007/3-540-61510-5_33
Download citation
DOI: https://doi.org/10.1007/3-540-61510-5_33
Published:
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-61510-1
Online ISBN: 978-3-540-68684-2
eBook Packages: Springer Book Archive