A CONNECTIONIST MODEL USING PROBABILITY DATA AS WEIGHTS AND PARAMETERS

Based on the concept of virtual lateral inhibition [1,2], a two layere d connectionist model is developed, and its properties explored. This model is called LIBRA/RX. The flow of activation in this model is desc ribed by a set of 3N ordinary nonlinear differential equations, where N is the number of nodes on the nets' upper level. The mathematical pr operties of the equations are explored, and in particular, the dynamic s of the net is demonstrated to be convergent in nearly all cases. The model has thus far been employed in the task of pattern recognition, and more recently in control tasks [3]. In the task of pattern recogni tion, the lower level or input nodes represent the possible features, and the upper level or output nodes represent the possible classes of patterns. This model uses the probabilities of the pattern classes giv en the features, and the features given the pattern classes as weights . The prior probabilities of the features and pattern classes also app ear as parameters-thus, no learning need be involved. Examples of the nets use in classifying patterns are presented and discussed.