A two class pattern recognition algorithm is introduced for learning the optimal separating function within a given finite dimensional linear space of functions. It is proved that the algorithm converges with probability one to the separating function minimizing the probability of misclassification over the given class. This stochastic gradient process is based on asymptotically unbiased estimates of the gradient vector of error probability. This method is proposed. to improve classification rules obtained by other methods.
|Number of pages||12|
|Journal||Probl Control Inf Theory|
|Publication status||Published - Jan 1 1976|
ASJC Scopus subject areas