Item

Abstract

The cortex is often described as a network processing information in the direction from sensory to motor areas. However, the structure of the cortex is asymmetrical only in the vertical direction, suggesting an input-output transformation between layers rather than between areas. This operation must be a very generally applicable one, since the plan of the cortex is basically the same everywhere. In an attempt to understand it, a skeleton cortex of only pyramidal cells is considered. They are characterized by a double dendritic expansion, an apical one in the first layer, which is considered as the input layer, and a basal one which receives excitation from the axon collaterals of other pyramidal cells. If pyramidal cells learn (perhaps by growing dendritic spines) to respond to frequent constellations of activity in their afferents, each will learn a property of the input (through its apical dendrites) provided that it was preceded by other properties sensed by neighbouring pyramidal cells (which influences it through its basal dendrites). Thus the pyramidal cells will code the input in terms of properties which have a tendency to follow each other. This will be a coding which reflects the causal structure of the world. Various uses of a network embodying the conditional probabilities of events in the input are described, including recognition of familiar sequences and prediction. The local variation of fiber patterns in the cerebral cortex of man, described as myeloarchitectonics, is interpreted as a macroscopical expression of the different statistics of the set of conditional probabilities linking the events represented by individual pyramidal cells in different areas (in different functional contexts).