Connectionist models and Bayesian inference
James L. McClelland · 2000
Abstract The study of what might be the rational or optimal response to a situation, given an intrinsically uncertain world, owes a great deal to the Reverend Bayes, and in contemporary times to a growing band of information processing theorists, cognitive scientists, and artificial intelligence researchers who have pressed extensions and applications of Bayesian ideas. Our most widely used optimal model within psychology may be the theory of the ideal observer within signal detection theory (Green and Swets, 1966). Within machine vision, the problem of seeing is often framed as one of selecting the most probable or ‘optimal’ interpretation of an input array, based on Bayesian calculations (e.g. Biilthoff and Yuille, 1996). Within cognitive psychology, models that address optimality spring in large measure from the recent work of John Anderson, who has used them to address a range of aspects of cognition, from perception to categorization to memory retrieval.