Optical neural network-architecture, design and models

An adaptive optical neural network (ONN) using inexpensive pocket size liquid crystal televisions was constructed by a group of graduate students in the electrooptics laboratory at the Pennsylvania State University. The major advantages of this optical neural architecture are its flexibility in operations and low construction cost. To test the performance of the ONN, several neural network models have been used. These models are Hopfield, orthogonal projection, multilevel recognition, interpattern association, unsupervised learning, and heteroassociation. The system design considerations and experimental demonstrations are given. Simple translation between English alphabets and Chinese characters have been demonstrated