Title :
Bounds for hidden units of simple recurrent networks
Author :
Ludik, Jacques ; Cloete, Ian
Author_Institution :
Dept. of Comput. Sci., Stellenbosch Univ., South Africa
Abstract :
We determine upper and lower bounds for the number of hidden units of Elman and Jordan architecture-specific recurrent threshold networks. The question of how many hidden units are enough to realize an arbitrary function is addressed not only in terms of the number of examples in general position, but also in terms of the distance between two distinct classes. We have also pointed out and corrected some mistakes in proofs for determining bounds for the number of hidden units of feedforward threshold networks
Keywords :
feedforward neural nets; recurrent neural nets; Elman/Jordan architecture-specific recurrent threshold networks; feedforward threshold networks; hidden units; lower bounds; simple recurrent networks; upper bounds; Africa; Computer architecture; Computer networks; Computer science; Euclidean distance; Hypercubes; Sun;
Conference_Titel :
Neural Networks, 1996., IEEE International Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-3210-5
DOI :
10.1109/ICNN.1996.548912
