On design and evaluation of tapped-delay neural network architectures

Abstract
We address pruning and evaluation ofTapped-Delay Neural Networks for the sunspot benchmarkseries. It is shown that the generalization abilityof the networks can be improved by pruning using theOptimal Brain Damage method of Le Cun, Denkerand Solla. A stop criterion for the pruning algorithmis formulated using a modified version of Akaike's FinalPrediction Error estimate. With the proposed stopcriterion the pruning scheme is shown to produce succesfularchitectures with a high yield.I. ...

This publication has 4 references indexed in Scilit: