Pattern association by lateral-alignment-free optical neural network

Tsumura Norimichi, Yusaku Fujii, Kazuyoshi Itoh, Yoshiki Ichioka

Osaka University

Optik 98 (1994)65-70.

Abstract
We propose an optical technique that solves partially the problem of alignment of optical neural networks. The synaptic weights are recorded on a special SLM that is located between input and output layers of optical artificial neurons. The SLM can accept the optical signals for changing the synaptic weights from both the input and output layers. The positions of synaptic weights to be recorded are automatically aligned by the positions of the neurons in the both layers. Results of an association experiment from Chinese characters to alphabets are presented.


Full paper
[PDF] (reprint)


Back