Title :
Fault-tolerant MLP networks for improved reliability and safety
Author :
Rughooputh, Harry C S ; Backory, Jay K.
Author_Institution :
Fac. of Eng., Mauritius Univ., Reduit, Mauritius
Abstract :
Presents a modified multi-layer perceptron (MLP) network called the modular redundant MLP (MR-MLP) network that shows improved reliability and safety. The standard MLP is first augmented by duplicating the hidden units and then the output module is duplicated m times to give an m MR-MLP. The outputs of the MR-MLP network are processed by voting circuits that produce the final outputs of the system together with a flag that indicates the validity of the output vector
Keywords :
multilayer perceptrons; redundancy; reliability; safety; fault-tolerant MLP networks; modular redundant MLP network; reliability; safety; voting circuits; Artificial neural networks; Biological neural networks; Circuit faults; Fault tolerance; Multilayer perceptrons; Redundancy; Reliability engineering; Robustness; Safety; Voting;
Conference_Titel :
Neural Networks, 1995. Proceedings., IEEE International Conference on
Print_ISBN :
0-7803-2768-3
DOI :
10.1109/ICNN.1995.487344