Ju-Seog Jang

Optical implementation of the Hopfield model for two-dimensional associative memory

Optical implementation of Hopfields neural network model [Proc. Natl. Acad. Sci. USA 79, 2554 (1982)] for two-dimensional associative memory is discussed. Two-state neuron elements are represented by a twisted nematic liquid-crystal optical switch array, and three-dimensional holographic interconnections are realized with these elements. Unipolar connections, created by adding a constant to bipolar interconnections and compensating them with an input-dependent thresholding operation, are realized. The 16- (4 x 4) neuron system model acts as a content-addressable associative memory with error-correction capability.

Programmable quadratic associative memory using holographic lenslet arrays

Programmable higher-order optical interconnections between two-dimensional arrays of neurons are achieved by using holographic lenslet arrays and spatial light modulators. Basic experimental results for two-dimensional quadratic associative memory are reported to demonstrate its feasibility.

OPTICAL IMPLEMENTATION OF QUADRATIC ASSOCIATIVE MEMORY WITH OUTER-PRODUCT STORAGE

Optical implementation of quadratic associative memory with outer-product storage is reported. Weighted N3 interconnections between neurons are realized with an optical vector–matrix multiplier and interconnection holograms.

Dynamic optical interconnections using holographic lenslet arrays for adaptive neural networks

Fully programmable higher order optical interconnections are described using holographic lenslet arrays and spatial light modulators. Adaptive neural net models can be implemented with this interconnection scheme. To demonstrate its feasibility, basic experiments conducted for adaptive neural network models are reported.

Implementation of the arbitrary free-space optical interconnections with equal path lengths by use of a layered structure.

We describe and demonstrate a scheme to implement arbitrary space-variant optical interconnections with equal path lengths in free-space, using a layered structure for a low gate count and high throughput. We discuss extending the technique to reduce the number of layers by performing small group interconnections on laterally separable locations.

Optical neural-net analog-to-digital converter

A parallel analog-to-digital converter with neuronlike elements is designed and optically implemented. Its operation principle is the simultaneous estimation of bit values for a given analog input. Architecture of the proposed analog-to-digital converter is simpler than that of an earlier one designed by the energy-minimization technique, and its digital output is independent of the initial state.