Astrocytes and slow learning in the formation of distal cortical associations

Abstract

An extension is proposed of the range of biologically inspired models for local computation architectures and learning in ANNs. The specific focus is on learning processes underlying the formation of connections providing distal access between local neural areas engaging in subsymbolic processing. A neural network architecture underlying the construction of distal cortical associations representing shared, specific contextual relevance between local regions is described. The model involves the evolutionary co-option of hippocampal neural structures to new functions and the adoption of a slow competitive learning process involving interaction between astrocytic and neural processes to enable identification of appropriate associations between cortical and motivational/emotional activity which proceed on different time scales.