Error-correcting algorithm for adapting a hyperelliptical decision surface for pattern recognition

An error-correction method of modifying the position and size of a hyperelliptical discriminant surface is described. The method uses an external teacher from which the only information required is the correct classification for each pattern in the training set. The algorithm may readily be implemented by simple analogue hardware. A digital-computer simulation is described, and results are presented both for the case in which matching between the machine and the teacher is possible, and for the case in which perfect matching is impossible.