Coenen, O. J.-M. D., Boucheny, C., Marchal, D., Arnold, M.P., Barbour, B., Ortigosa, E., Agís, R. and Ros, E. Spiking cerebellar model for adaptive control. Forum of European Neuroscience Abstracts, 2004

Sony CSL authors: Michele Bezzi, Christian Boucheny, Olivier Coenen, David Marchal

Abstract

A spiking cerebellum model and motor control subsystems are integrated to form a platform for future investigations in learning and context modulation of movement control. The elements of the model are as follows: A spiking cerebellar model generates anticipation of perception and action whereas a feedforward motor controller maps desired hand positions in spatial coordinates to the activation ratios required to control a 6-muscle simulated human arm model. The cerebellar model is composed of four types of interacting spiking neural networks: granular layers, Purkinje layers, cerebellar nuclei and inferior olives. Sensory inputs are pre-processed by the granular layers with adaptation in the mossy fiber - granule cell synapses. Supervised learning at the parallel fiber - Purkinje cell and mossy fiber - cerebellar nuclei synapses is mediated by the inferior olives. The cerebellar nuclei project to a motor controller. This module learns the mapping of the arm through experience and drives the arm movement according to the cerebellar signals. The combined system is demonstrated on a test task inspired by the pong videogame. In this task, the cerebellum predicts the trajectory of the ball in different contexts, and the simulated arm moves the racket accordingly.

Keywords: Cerebellum, control, spiking computation, modeling, theory

BibTeX entry

@INPROCEEDINGS { coenen:04a, AUTHOR="Coenen, O. J.-M. D. and Boucheny, C. and Marchal, D. and Arnold, M.P. and Barbour, B. and Ortigosa, E. and Agís, R. and Ros, E.", BOOKTITLE="Forum of European Neuroscience Abstracts", TITLE="Spiking cerebellar model for adaptive control", YEAR="2004", }