Abstract

A novel type of quantized learning rule with unipolar binary weights that is useful for the optical implementation of neural networks is reported. An input-dependent thresholding operation is also proposed to remove the unwanted effects that are due to the insufficient contrast ratio of spatial light modulators as synaptic connection devices. Moreover, we experimentally demonstrate the recognition of 26 characters of the alphabet by using the single set of an optoelectronic three-layered network.

© 1990 Optical Society of America

Influence of interconnection weight discretization and noise in an optoelectronic neural network

A. Von Lehmen, E. G. Paek, P. F. Liao, A. Marrakchi, and J. S. Patel
Opt. Lett. 14(17) 928-930 (1989)

Optical outer-product learning in a neural network using optically stimulable phosphor

Fumihiko Itoh, Ken-ichi Kitayama, and Yasuaki Tamura
Opt. Lett. 15(15) 860-862 (1990)

Optical implementation of large-scale neural networks using a time-division-multiplexing technique

Masaya Oita, Jun Ohta, Shuichi Tai, and Kazuo Kyuma
Opt. Lett. 15(4) 227-229 (1990)

Neural network recognition and classification of aerosol particle distributions measured with a two-spot laser velocimeter

Eugene Yee and Jim Ho
Appl. Opt. 29(19) 2929-2938 (1990)

Unipolar interpattern association neural networks

Chii-Maw Uang, Guowen Lu, and Francis T. S. Yu
Opt. Lett. 19(1) 52-54 (1994)