Optical associative memory with invariances

IEEE International Workshop on Cellular Neural Networks and their ApplicationsPublished 4 December 2002

Guoguang Yang

DOI10.1109/cnna.1990.207534

Citations6

TL;DR

A single-slab second order neural network model with scale and translation invariance based on the backpropagation learning rule by using group theory to impose the invariances to the network is proposed and the performance of network is improved drastically.

Abstract

A single-slab second order neural network model with scale and translation invariances is proposed. This is based on the backpropagation learning rule by using group theory to impose the invariances to the network. The results show that full range translation invariance and a limited range of scale invariance are realisable. The performance of outer-product model with invariance is analysed. Then an inner-product associative memory model with translation invariance is proposed. A strictly increasing nonlinear operation which is cooperated with a sign function is chosen to guarantee the correct recall convergence of the network without iteration. The performance of network is improved drastically. These two models can be implemented by simple optical systems with parallel processing capability.>

Keywords

Computer Science

The MIT Press eBooksParallel Distributed Processing

15,227 Citations1986David E. Rumelhart, James L. McClelland +1 more

Introduction to Fourier optics

10,172 Citations1968Joseph W. Goodman

Applied OpticsOptical implementation of the Hopfield model

583 Citations1985Nabil H. Farhat, Demetri Psaltis +2 more

Numerical and experimental results presented show that the approach is capable of introducing accuracy and robustness to optical processing while maintaining the traditional advantages of optics, namely, parallelism and massive interconnection capability.

View PDF

Optical EngineeringShift-Invariant Optical Associative Memories

25 Citations1987Demetri Psaltis, John Hong

It is shown that full shift invariance cannot be achieved with systems that employ only linear interconnections to store the associations.

View PDF

Optics LettersOptoelectronic chip implementation of a quadratic associative memory

14 Citations1990A. Von Lehmen, Eung Gi Paek +3 more

This work describes an optoelectronic implementation of a second-order, or quadratic, associative memory based on an inner-product representation, which allows for a compact implementation using an optical chip design.

Electronics LettersLearning by diffusion for multilayer perceptron

10 Citations1989Richard Hoptroff, Trevor J. Hall

A method of nonlinear optimisation based on simulating diffusion processes is discussed and applied as a self-organisation or learning tool for the multilayer perceptron neutral network and results indicate significant improvements over conventional learning methods.

Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIEA Polarization Based Optical Connectionist Machine

2 Citations1989Mike Kranzdorf, Kristina M. Johnson +3 more

An optoelectronic connectionist module with modifiable inputs and weight matrices that performs associative memory operations is demonstrated that will allow the extension to many connectionist models.