Review of mapping neural networks in classification systems

Abstract

An old approach used in classification systems has been revived by the discovery of new algorithms which use this approach to approximate complex non-linear functions. This approach, known as neurocomputing, classifies input data in parallel. Neurocomputers should be exploited for their parallelism for use in problems that are currently being solved by serial classification systems such as know/edged-based and pattern recognition systems. This paper reviews single­-layer and multi-layer feed-forward neural networks, termed mapping neural networks since they approximate a function or mapping, and describes their functionality in classification systems. They have been successfully used as knowledge­-bases in expert systems, and have been shown to be functionally equivalent to several conventional classifiers used in pattern recognition.