A QUANTIZED OPTICAL NEURAL-NETWORK WITH ONLINE LEARNING

Authors
PIGNON D HALL TJ XU LQ PARMITER F DAINTY P RUSS KB CHERRY SR ROBERTS NC PREWETT PD
Citation
D. Pignon et al., A QUANTIZED OPTICAL NEURAL-NETWORK WITH ONLINE LEARNING, International journal of optoelectronics, 10(2), 1996, pp. 79-88
Citations number
12
Categorie Soggetti
Engineering, Eletrical & Electronic",Optics
Journal title
International journal of optoelectronics → ACNP
ISSN journal
09525432
Volume
10
Issue
2
Year of publication
1996
Pages
79 - 88
Database
ISI
SICI code
0952-5432(1996)10:2<79:AQONWO>2.0.ZU;2-E
Abstract
An optoelectronic hardware system that performs a matrix-matrix multip lication is presented. The system is based on electrically addressable spatial light modulators for switching and computer generated hologra ms for connection routeing. The matrix-matrix multiplier is used to im plement a quantized neural network. A novel algorithm for on-line lear ning for networks with quantized weights and hard thresholds based on function smoothing and stochastic approximation is used.