Modular architecture for Hopfield network and distance based training algorithm for pattern association

Authors:
Ming-Jung Seow;Vijayan K. Asari
Affiliations:
Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA;Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA
Venue:
Neural, Parallel & Scientific Computations
Year:
2002

Citing 7
Cited 0

The capacity of the Hopfield associative memory

IEEE Transactions on Information Theory
A competitive modular connectionist architecture

NIPS-3 Proceedings of the 1990 conference on Advances in neural information processing systems 3
HAVNET: a new neural network architecture for pattern recognition

Neural Networks
High Storage Capacity Architecture for Pattern Recognition Using an Array of Hopfield Neural Networks

AIPR '01 Proceedings of the 30th on Applied Imagery Pattern Recognition Workshop
Multiobjective genetic algorithm partitioning for hierarchical learning of high-dimensional pattern spaces: a learning-follows-decomposition strategy

IEEE Transactions on Neural Networks
Modular recurrent neural networks for Mandarin syllable recognition

IEEE Transactions on Neural Networks
Block-based neural networks

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper, a two-dimensional modular architecture for Hopfield neural network and a distance based training algorithm that improves the storage capacity and reduces the structural complexity of Hopfield neural network are presented. Our approach involve dividing a N×M network into (N×M)/(n×m) modules of size n×m with each module functioning independently as a sub-network in conjunction with the inputs from neighboring modules. In this technique, a divide and conquer approach, which permits us to solve a complex computational task by dividing it into simpler subtasks and then combining their individual solutions is utilized. This use of task- decomposition provides a modular structure with several advantages, viz. more reasonable generalization, more intelligible and useful representations, and more efficient use of computational hardware. The performance of the proposed technique is evaluated by applying it into various character images. It has been observed that the network exhibits faster convergence characteristics and is capable of reproducing learned patterns successfully from noisy and partial data.