We propose a new optical method for the generalized Hebbian-rule in the optical neural-network systems. In this method, the weight is recorded on an optically addressed spatial light modulator (OASLM). The modification of the weight is automatically achieved on the OASLM by the light signals from the post- and pre-synapse artificial neurons. This architecture makes it easy to align the optical system. An optically addressed FLC-SLM is used as the OASLM. A binary correlation-learning in the binary generalized Hebbian-rule is adopted in the present method. To improve the binary correlation-learning, we propose a learnable neural network for the sparse encoding. We verify the principle and efficiency of the proposed method through a preliminary experiment.
[1]
Shun-ichi Amari,et al.
Characteristics of sparsely encoded associative memory
,
1989,
Neural Networks.
[2]
K Wagner,et al.
Multilayer optical learning networks.
,
1987,
Applied optics.
[3]
Geoffrey E. Hinton,et al.
A general framework for parallel distributed processing
,
1986
.
[4]
H Toyoda,et al.
Optical associatron: a simple model for optical associative memory.
,
1989,
Applied optics.
[5]
M Toyama,et al.
Unevenness of illuminance caused by gradient-index fiber arrays.
,
1980,
Applied optics.