Reinforcement-Driven Dimensionality Reduction- A Model For Information Processing In The Basal Ganglia

Izhar Bar-Gad, Joshua A. Goldberg, Hagai Bergman, Gali Havazelet-Heimer, Eytan Ruppin

Research output: Contribution to journalArticlepeer-review


Although anatomical studies of the basal ganglia show the existence of extensive convergence and lateral inhibitory connections, physiological studies failed to show correlated neural activity or lateral interaction in these nuclei. These seemingly contradictory results could be explained with a model in which the basal ganglia reduce the dimensionality of cortical information using optimal extraction methods. Simulations of this model predict a transient change in the efficacy of the feed-forward and lateral synapses following changes in reinforcement signal, causing an increase in correlated firing rates. This process ultimately restores the steady-state situation with diminished efficacy of lateral inhibition and no correlation of firing. Our experimental results confirm the model's predictions: rate correlations show a drastic decrease between the input stage (cortex) and output stage (pallidum). Moreover, preliminary analysis revealed that pallidal correlations show a transient increase following discrepancies between the animals predictions and reality. We therefore propose that by using a reinforcement-driven dimensionality reduction process the basal ganglia achieve efficient extraction of cortical salient information that may then be used by the frontal cortex for execution and planning of forthcoming actions.

Original languageEnglish
Pages (from-to)305-320
Number of pages16
JournalJournal of Basic and Clinical Physiology and Pharmacology
Issue number4
StatePublished - 2000


  • MPTP
  • Parkinson's disease
  • cross-correlation
  • dopamine
  • globus pallidus
  • monkey


Dive into the research topics of 'Reinforcement-Driven Dimensionality Reduction- A Model For Information Processing In The Basal Ganglia'. Together they form a unique fingerprint.

Cite this