Abstract
An optical neural network system with learning capabilities and a 2-D architecture is proposed. The 2-D architecture is implemented by a multiple imaging system of a Selfoc microlens array, which enables us to develop a compact optical neural network system. Synaptic weights of neurons are memorized in the Pockels readout optical modulator, and the system obtains the learning capabilities. Experimental verification of multiple imaging and learning is presented. In the experiments, the neural network system stores the three patterns by an orthogonal learning method. The performance of the optical neural network system is estimated in terms of a point spread function of the Selfoc microlens.
