Title :
Recurrent nets and signal classification
Author :
Lugtenburg, S.P. ; Barnard, E.
Author_Institution :
Dept. of Electron. & Comput. Eng., Pretoria Univ., South Africa
fDate :
8/30/1991 12:00:00 AM
Abstract :
The feature space map used for analysis of feedforward backpropagation neural networks is extended to recurrent neural networks. The existence of difficulties in signal classification is shown. It is finally investigated whether this difficulty means that recurrent nets are not useful in signal-recognition tasks; the authors conclude that this is not so
Keywords :
neural nets; signal processing; feature space map; recurrent neural networks; signal classification; Backpropagation; Computer networks; Feature extraction; Feedforward neural networks; Geometry; Handwriting recognition; Neural networks; Pattern classification; Recurrent neural networks; Robustness;
Conference_Titel :
Communications and Signal Processing, 1991. COMSIG 1991 Proceedings., South African Symposium on
Conference_Location :
Pretoria
Print_ISBN :
0-7803-0040-8
DOI :
10.1109/COMSIG.1991.278240
