A new neural network architecture for efficient close proximity match of large databases

A new paradigm of multiple neural networks organized in a hierarchical structure is used to perform close proximity matches in large databases. The basic unit of this hierarchical structure is a modified Hopfield neural network with a new off-line training scheme. This scheme is adopted as it provides a quick learning property with no spurious states. The neural networks are not trained by the actual data, rather signatures which are obtained by superimposed coding on partitions of the input data. This allows the networks to be smaller sizes and the hierarchical structure allows a small sample of input data for training the networks. The paradigm is tested on two extremely opposite databases, viz., library retrieval systems and protein databases. The experimental results are reported to corroborate that high level of precision can be obtained for efficient recall of inexact queries using the proposed neural network