Title :
Learning-based on-line testing in feedforward neural networks
Author :
Kamiura, Naotake ; Yamato, Kazuharu ; Isokawa, Teijiro ; Matsui, Nobuyuki
Author_Institution :
Dept. of Comput. Eng., Himeji Inst. of Technol., Japan
Abstract :
Learning-based on-line testing in feedforward neural networks (NNs) is discussed. After the convergence of the ordinary learning, the re-learning employing two sigmoid activation functions per neuron in the last layer of the NN is made. It sets up the range of erroneous potentials produced from the last layer, and enables us to detect faults without extra hardware.
Keywords :
backpropagation; convergence of numerical methods; fault simulation; feedforward neural nets; neural chips; pattern recognition; arbitrary learning patterns; convergence; erroneous potentials range; fault models; feedforward neural networks; learning-based on-line testing; link survival probability; parity matrix; pattern recognition problem; retraining; sigmoid activation functions; standard backpropagation algorithm; Circuit faults; Circuit testing; Computer networks; Fault detection; Feedforward neural networks; Information science; Intelligent networks; Neural network hardware; Neural networks; Neurons;
Conference_Titel :
On-Line Testing Workshop, 2002. Proceedings of the Eighth IEEE International
Print_ISBN :
0-7695-1641-6
DOI :
10.1109/OLT.2002.1030206
