Reinforcement Neuronal Signals and Dimensionality Reduction in the Cortico-Basal Ganglia System

Abstract

Although anatomical studies of the basal ganglia show extensive convergence and lateral inhibitory connections (Bolam et al., 2000), physiological studies failed to show correlated pallidal activity (Nini et al., 1995;Raz et al., 2000). These seemingly contradictory results can be explained within a model, which postulates that the basal ganglia reduce the dimensionality of cortical information using optimal extraction methods (Bar-Gad et al., 2000). Such compression can be controlled by the reinforcement neuronal systems of the basal ganglia, i.e., the midbrain dopaminergic neurons (Schultz, 1998) and the cholinergic interneurons of the striatum (Graybiel et al., 1994). Previous studies have shown that midbrain dopaminergic neurons respond to mismatch between the animal’s predictions and reality. Thus, dopaminergic neurons respond to primary rewards only when the reward occurs unpredictably (Hollerman and Schultz, 1998). In a trained animal, the dopaminergic neurons increase their discharge with the earliest reward-predicting stimulus, and reduce their discharge when this prediction fails to occur (Mirenowicz and Schultz, 1994). The tonically active neurons (TANs) of the striatum (presumably the cholinergic interneurons of this structure (Wilson et al., 1990;Aosaki et al., 1995)) show responses of a similar nature (Graybiel et al., 1994;Ravel et al., 1999), although with reverse polarity.