Title: DESIGNER NETWORKS FOR TIME SERIES PROCESSING
Authors: C. Svarer, L. K. Hansen, J. Larsen and C. E. Rasmussen
CONNECT, Electronics Institute B349
Technical University of Denmark
DK-2800 Lyngby, Denmark
+45 4593 4207 Fax: +45 4288 0117
emails: claus,lkhansen,jlarsen@ei.dtu.dk
Design of Tapped-Delay Neural Networks for time series processing is
studied. We corroborate recent results on pruning using the
Optimal Brain Damage method of Le Cun, Denker and Solla. In particular
we show that the generalization ability of these networks can be
improved for applications within
inverse modeling and chaotic system identification.
We use a stop criterion for pruning based on a modified
version of Akaike's Final Prediction Error estimate.
In C.A. Kamm, G.M. Kuhn, B. Yoon, R. Chellappa, S.Y. Kung: Proceedings of the
III IEEE Workshop on Neural Networks for Signal Processing, IEEE Press, 1993,
pp. 78--87.