Programmable Optical Quadratic Neural Networks.

摘要

Analytical and experimental investigations of optical quadratic neural networks have been conducted. Results reported include: Development of a polarization-based outer product processor implementing a quadratic neural network; implementation of a photorefractive crystal based quadratic neural network utilizing feedback for learning; derivation of a statistical parameter which can be used to completely characterize the behavior of linear and higher order Hopfield associative memories; and development of a method to reduce the number of interconnections in associative memories while retaining good performance.