An Easily Mechanized Scheme for an Adaptive Pattern Recognizer

Adaptive pattern recognizers incorporate approximations to the class probability distributions that improve as a result of cumulative information derived from known inputs. A simple method is outlined in which a stored sample constitutes a set of points in observation-space at each of which the local value of a class's distribution can be estimated cumulatively as known class-members become available. The membership of the sample representing each class is modified to fill gaps and to adjust to the relative spread of the distribution. Only a minimum of computation, mostly simple counting, is required.