Two Viewpoints of k-Tuple Pattern Recognition

This paper presents two viewpoints of the k-tuple pattern recognition scheme proposed by Browning and Bledsoe. The first shows that k-tuple pattern recognition is a statistical approximation technique. In effect, the recognition is accomplished by approximating a higher order probability distribution by use of the first-order distributions. Using this viewpoint, and Lewis' measure of characteristic selection, several alternative approximations are offered. The second viewpoint is that recognition is a special case, or subclass, of a Ф learning machine. It can be shown that if the input pattern vector X is first processed by a Ф-processor (in this case a kth order polynomial) and then certain terms discarded, the resulting learning machine is identical to a k-tuple pattern recognition machine.