PERCEPTUAL CATEGORIZATION - CONNECTIONIST MODELING AND DECISION RULES

Although it is currently popular to model human associative learning u sing connectionist networks, the mechanism by which their output activ ations are converted to probabilities of response has received relativ ely little attention. Several possible models of this decision process are considered here, including a simple ratio rule, a simple differen ce rule, their exponential versions, and a winner-take-all network. Tw o categorization experiments that attempt to dissociate these models a re reported. Analogues of the experiments were presented to a single-l ayer, feed-forward, delta-rule network. Only the exponential ratio rul e and the winner-take-all architecture, acting on the networks' output activations that corresponded to responses available on test, were ca pable of fully predicting the mean response results. In addition, unli ke the exponential ratio rule, the winner-take-all model has the poten tial to predict latencies. Further studies will be required to determi ne whether latencies produced under more stringent conditions conform to the model's predictions.