A Stochastic Search Algorithm to Optimize an N-tuple Classifier by Selecting Its Inputs

The N-tuple method [4] is a statistical pattern recognition method, which decomposes a given pattern into several sets of n points, termed “N tuples”. The input connection mapping of the N-tuple classifier determines the sampling and defines the locations of the pattern matrix. Realizing the fact that the classification performance of the N-tuple classifier is highly dependant on the actual subset of the input bits probed [3][7], we have introduced an approach based on a Reward and Punishment (RnP) scheme to select input mappings of the classifier. We termed the classes with high error rates as critical classes. Different groups of tuples have been formed for different classes. The strategy was to employ more number of tuples to a critical class-group than an easily distinguishable class. In order to illustrate the capabilities of the RnP based measure the task of recognizing hand-written digits from NIST [10] database has been chosen.