Storing covariance with nonlinearly interacting neurons

Area of Science:

• Computational Neuroscience
• Neuroscience
• Biophysics

Background:

• The Hartline-Ratliff model describes neuronal interactions in the Limulus retina.
• Previous work probabilistically analyzed a time-dependent, nonlinear model of neuronal interaction.

Purpose of the Study:

• To demonstrate that the nonlinear neuronal model is a generalization of the Hartline-Ratliff model.
• To investigate the role of neuronal membrane potentials and firing rates in information storage.
• To propose a mechanism for synaptic plasticity in cerebellar motor learning.

Main Methods:

• Probabilistic analysis of a time-dependent, nonlinear neuronal interaction model.
• Coupling equations for neuronal membrane potential means and covariances through average firing rates.
• Treating motor learning as a problem of covariance storage.

Main Results:

• The model generalizes the Hartline-Ratliff model.
• Average firing rates control the selective storage and retrieval of covariance information.
• Synaptic plasticity in the cerebellar cortex is predicted to be proportional to the covariance between parallel fiber and climbing fiber discharges.

Conclusions:

• The proposed model offers a novel perspective on neuronal information processing.
• The prediction for synaptic plasticity necessitates both facilitation and depression at the same synapse.
• This framework provides a basis for understanding motor learning mechanisms.