LEARNING BY SELECTION IN THE TRION MODEL OF CORTICAL ORGANIZATION

The basic issue of whether mammalian learning in cortex proceeds via a selection principle, as stressed by Edelman, versus an instructional one is of major importance. We present here a realization of selection learning in the trion model, which is based on the Mountcastle columnar organizational principle of cortex. We suggest that mammalian cortex starts out with an a priori connectivity between minicolumns that is highly structured in time and in space, competing between excitation and *inhibition. This provides a ''naive'' repertoire of spatial-temporal firing patterns that stimuli and internal processing map onto. These patterns can be learned with small modifications to the connectivity strengths determined by a Hebbian learning rule. As various patterns are learned, the repertoire changes somewhat in order to respond properly to various stimuli, but the majority of all possible stimuli still map onto spatial-temporal firing patterns of the original repertoire. In order to show that the example presented here is showing true selectivity and is not an artifact of more stimuli evolving into the learned pattern, we develop a selectivity measure. We suggest that some form of instructional learning (in which connectivities are finely tuned) is present for difficult tasks requiring many trials, whereas very rapid learning involves selectional learning. Both types of learning must be considered to understand behavior.