Abstract

The idea of n-tuple sampling as a basis for pattern recognition, as proposed by Bledsoe and Browning (1959), remains a viable approach to a range of pattern classification tasks especially where speed of learning is of importance. The formal relationship between n-tuple neural networks and more mainstream network paradigms, such as radial basis function networks, and classical nonparametric pattern classifiers, such as kernel estimation, is considered, and it is described how the classic n-tuple recogniser and the n-tuple regression network form differing approximations in the classification process