Pattern Recognition with Three Added Requirements

A number of additional constraints are added to the usual problem of forming a pattern recognition scheme using adaptive techniques. These require that the system formed be 1) capable of adding a new class without requiring excessive change of the original structure, 2) able to handle sequential patterns as generated by events which occur over a period of time, and 3) able to predict parts of the sequential pattern. A particular scheme is simulated on a large digital computer and tested on problems in speaker identification, speech recognition, and electroencephalogram recognition.