Title :
Influence on ANNs fault tolerance of binary errors introduced during training
Author :
Assoum, Ammar ; Geagea, Maïkel ; Velazco, Raoul
Author_Institution :
Fac. of Sci. III, Lebanese Univ., Tripoli, Lebanon
Abstract :
This paper presents the effect of binary errors on artificial neural networks during the training phase and the fault tolerance of ANN. The tested network implements a problem (recognition of digits and letters) and is trained in the presence of binary errors. A significant improvement is obtained with increase in the number of perturbations during the training phase and recognition rate during the generalization phase.
Keywords :
fault tolerance; generalisation (artificial intelligence); learning (artificial intelligence); neural nets; perturbation techniques; artificial neural network; binary error; digit recognition; fault tolerance; generalization phase; network test; perturbation; recognition rate; training phase; Application software; Artificial neural networks; Circuit simulation; Fault tolerance; Laboratories; Neural network hardware; Neural networks; Neurons; Single event upset; Testing;
Conference_Titel :
Information and Communication Technologies: From Theory to Applications, 2004. Proceedings. 2004 International Conference on
Print_ISBN :
0-7803-8482-2
DOI :
10.1109/ICTTA.2004.1307818
