Abstract
Neural networks are known to be massively interconnected and to process information in parallel. Parallelism and massive interconnection capability are recognized strengths of optics. Increasing attention is therefore being given to the optical modeling of neural processing. Optoelectronic implementation and performance evaluation of a neural net consisting of thirty-two neurons based on the Hopfield model which is suitable for use with binary 1-D input vectors was reported recently.1 In this paper we discuss extension of the model to 2-D formats. The underlying principle and detail of two schemes for optical implementation of the required 4-D memory matrix (synaptic mask) are given. These are based on the use of either spatial multiplexing or spatial frequency multiplexing. Practical design considerations are covered.
© 1985 Optical Society of America
PDF ArticleMore Like This
K. S. Lee and Nabil H. Farhat
WJ35 OSA Annual Meeting (FIO) 1985
Nabil Farhat and Demetri Psaltis
WB3 Optical Computing (IP) 1985
M. M. Mirsalehi and T. K. Gaylord
WB1 Optical Computing (IP) 1985