PROBABILISTIC WINNER-TAKE-ALL LEARNING ALGORITHM FOR RADIAL-BASIS-FUNCTION NEURAL CLASSIFIERS

This paper proposes a new adaptive competitive learning algorithm call ed ''the probabilistic winner-take-all.'' The algorithm is based on a learning scheme developed by Agrawala within the statistical pattern r ecognition literature (Agrawala 1970). Its name stems from the fact th at for a given input pattern once each competitor computes the probabi lity of being the one that generated this pattern, the computed probab ilities are utilized to probabilistically choose a winner. Then, only this winner is permitted to learn. The learning rule of the algorithm is derived for three different cases. Its properties are discussed and compared to those of two other competitive learning algorithms, namel y the standard winner-take-all and the maximum-likelihood soft competi tion. Experimental comparison is also given. When all three algorithms are used to train the hidden layer of radial-basis-function classifie rs, experiments indicate that classifiers trained with the probabilist ic winner-take-all outperform those trained with the other two algorit hms.