Title :
Backpropagation learning using positive weights for multilayer optoelectronic neural networks
Author :
Bradley, W.M.D. ; Mears, R.J.
Author_Institution :
Dept. of Eng., Cambridge Univ., UK
Abstract :
The use of inhibitory neurons, together with a modified learning algorithm, allows error backpropagation learning that competes with standard momentum-aided backpropagation using positive weights only. A multilayer optoelectronic architecture with in situ learning is proposed.
Keywords :
backpropagation; multilayer perceptrons; optical neural nets; parallel algorithms; parallel architectures; backpropagation learning; error backpropagation learning; in situ learning; inhibitory neurons; modified learning algorithm; momentum-aided backpropagation; multilayer optoelectronic architecture; multilayer optoelectronic neural networks; positive weights; Backpropagation algorithms; Multi-layer neural network; Neural networks; Neurons; Optical computing; Optical fiber networks; Optical network units; Optical saturation; Smart pixels; Telephony;
Conference_Titel :
Lasers and Electro-Optics Society Annual Meeting, 1996. LEOS 96., IEEE
Conference_Location :
Boston, MA, USA
Print_ISBN :
0-7803-3160-5
DOI :
10.1109/LEOS.1996.565248
