Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

Symposium on Perspectives in Computational Neuroscience

Friday, 01.08.2014 09 s.t.

At the edge of chaos: how granular-layer dynamics can provide the basis for cerebellar function

by Dr. Christian Rössert
from The University of Sheffield, Sheffield, UK

Contact person: Fred Wolf

Location

Ludwig Prandtl lecture hall

Abstract

The cerebellum plays an important role in motor control and motor learning and therefore is essential for everyday tasks in animals and humans. While the cerebellum itself does not initiate movements it strongly contributes to their refinement and correction. Subsequently cerebellar dysfunction leads to erratic, uncoordinated, or incorrectly timed motion. A unifying theory that can explain many aspects of cerebellar function such as learning of internal models, state estimation, (sensory) signal processing and motor control is the highly versatile adaptive-filter model of the original Marr­Albus theoretical framework. The basic function of the adaptive-filter model involves two key features that are very congruent with the cerebellar cortical architecture: in the granular layer the input signals are supposed to be expanded and recoded to provide a foundation from which the Purkinje cells synthesise output filters to implement specific behavioural signals. While many aspects of the adaptive-filter model have been shown to be in agreement with electrophysiological findings, e.g. symmetrical learning at Purkinje cell synapses, the important general mechanism of expansion-recoding has not yet been identified. In this talk I will give a general introduction to cerebellar- function, architecture and theoretical models and will focus on the hypothesis that random recurrent inhibition in the granular layer could indeed provide the necessary input signal separation and lengthening as required for expansion-recoding. I further show that this proposed mechanism is very similar to the well-established machine learning approach of reservoir computing and that the quality of filter construction is best if the network is close to the edge of random chaotic behaviour. In conclusion I will provide an outlook on important potential multi-scale modelling studies ranging from cellular ion-channels to cerebellar-inspired robotic control.

back to overview