Perceptual learning has received enhanced interest during the last years bo
th from theoreticians and experimentalists. Recent experimental results rev
eal that mechanisms underlying perceptual learning are more complex than pr
eviously expected, thereby ruling out any explanations based on simple neur
al network models. These findings do not represent an insignificant excepti
on to the rule but are evidence that present models fail to reflect some im
portant characteristics of the learning process. A new model is introduced
that is able to overcome problems of classical neural networks and that mig
ht be viewed as a hybrid between supervised and unsupervised learning. (C)
1999 The Optical Society of America. [S1070-9762(99)01209-9].