PARALLEL COMPUTATION IN PATTERN RECOGNITION

Abstract

One of the rather curious factors hindering any coordinated attack on the problems of pattern recognition is the difficulty experienced in reaching agreement as to what is meant by a pattern. The need to give some precision to the definition of the word is felt while attempting to devise a pattern recognition system. It is convenient, if not essential, to regard any pattern recognition system as including in its early stages some form of input device that will represent the input as an ordered set of numbers. The actual method used, although it may require the application of very sophisticated techniques and a depth of understanding of the properties of the input that may not be readily available, is irrelevant to the present argument. The important assumption is that the input device will not destroy any part of the input information that is necessary for the pattern to be recognized by analysis of the ordered number set. It is a matter for discussion as to whether the character of a pattern recognition process is such that a parallel representation is likely to be the most fruitful. It has been opinioned that too much notice has been taken of physiological neural parallel circuitry, which might well be designed so as to allow circuit redundancy and regeneration rather than as a maximally efficient logical circuit. This chapter discusses the parallel computation in pattern recognition.