64-channel correlator implementing a Kohonen-like neural network for handwritten-digit recognition

Abstract

We present an optical implementation of an improved version of the Kohonen map neural network applied to the recognition of handwritten digits taken from a postal code database. Improvements result from the introduction of supervision during the learning stage, a technique that also simplifies the map layer labeling. The experimental implementation is based on a frequency-multiplexed raster computer-generated hologram used to realize the required N(4) interconnection. The setup is shown to be equivalent to a 64-channel correlator. Computer simulations are used to study various detection and classification procedures. The results of the optical experiments, obtained with binary phase computer-generated holograms, are presented and shown to be in excellent agreement with the simulations.

High-capacity photorefractive neural network implementing a Kohonen topological map

“…Therefore the use of free-space optics may be a good way to implement such systems because of the high connectivity and massive parallelism of such a system. Several optical neural networks have already been built, [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] but few of them have actually achieved large capacities. [13][14][15][16] So our aim was to demonstrate the feasibility of such a system and to find the problems that are caused by high capacities.…”

Automatic Classification of Images with a Photorefractive Crystal

“…This type of systems is based on the interconnection of a large number of elementary units. This leads rather naturally to the idea of using optical interconnections to implement these neural networks, which allows to gain profit of the massive parallelism of optics [3][4][5]. Among all the optical processes, volume holographic interconnects are an interesting way of research because of their large potential capacity [6].…”