Realization of CNNs by optical parallel processing with spatial light valves

Abstract

Proposes an optical realization of cellular neural networks (CNNs). Any implementation of CNNs must calculate correlations of templates and input or output states of neurons. Therefore the system described is based on an optical special purpose processor which is perfectly suited for the computation of real time correlations. The processor is made of lenses and electronically addressed liquid crystal light valves which permit real time modifications of inputs and templates. Large neural nets and templates which are not restricted in their connectivity can be realized with massive optical parallel processing.<>