Fast convergence of cerebellar learning (Luque et al. 2015)

 Download zip file 
Help downloading and running models
The cerebellum is known to play a critical role in learning relevant patterns of activity for adaptive motor control, but the underlying network mechanisms are only partly understood. The classical long-term synaptic plasticity between parallel fibers (PFs) and Purkinje cells (PCs), which is driven by the inferior olive (IO), can only account for limited aspects of learning. Recently, the role of additional forms of plasticity in the granular layer, molecular layer and deep cerebellar nuclei (DCN) has been considered. In particular, learning at DCN synapses allows for generalization, but convergence to a stable state requires hundreds of repetitions. In this paper we have explored the putative role of the IO-DCN connection by endowing it with adaptable weights and exploring its implications in a closed-loop robotic manipulation task. Our results show that IO-DCN plasticity accelerates convergence of learning by up to two orders of magnitude without conflicting with the generalization properties conferred by DCN plasticity. Thus, this model suggests that multiple distributed learning mechanisms provide a key for explaining the complex properties of procedural learning and open up new experimental questions for synaptic plasticity in the cerebellar network.
1 . Luque NR, Garrido JA, Carrillo RR, D'Angelo E, Ros E (2014) Fast convergence of learning requires plasticity between inferior olive and deep cerebellar nuclei in a manipulation task: a closed-loop robotic simulation. Front Comput Neurosci 8:97 [PubMed]
Model Information (Click on a link to find other models with that property)
Model Type: Realistic Network;
Brain Region(s)/Organism: Cerebellum;
Cell Type(s):
Gap Junctions:
Simulation Environment: Simulink;
Model Concept(s): STDP;
Implementer(s): Garrido, Jesus A [jesus.garrido at]; Luque, Niceto R. [nluque at];
%ACCEL Compute manipulator forward dynamics
% Returns a vector of joint accelerations that result from applying the 
% actuator TORQUE to the manipulator ROBOT in state Q and QD.
% Uses the method 1 of Walker and Orin to compute the forward dynamics.
% The accelerations of the coordinates are obtained first 
% with the method of Walker-Orin and, later,it is joining to obtain speed and position.  

% This form is useful for simulation of manipulator dynamics, in
% conjunction with a numerical integration function.
% Walker and Orin is a numerical method used to obtain the acceleration of the
% articular coordinates from the torque vector.For it, Newton-Euler's
% algorithm uses when articular aceleration is zero
% B= 0+H(q,q')+C(q); tau=D(q)q''+B; q''=inv(D(q))[tau-B]

% See also: RNE, ROBOT, ODE45.

% 4/99 add object support
% 1/02 copy rne code from inertia.m to here for speed
% % General cleanup of code: help comments, see also, copyright, remnant dh/dyn
% references, clarification of functions.
%   1999 Peter I. Corke
%   2007 Niceto Luque Sola
function qdd = accel(robot, Q, qd, torque)
	n = robot.n;

	if nargin == 2,
	        q = Q(1:n);
		qd = Q(n+1:2*n);
		torque = Q(2*n+1:3*n);
		q = Q;
		if length(q) == robot.n,
			q = q(:);
			qd = qd(:);

	%   compute current manipulator inertia
	%   torques resulting from unit acceleration of each joint with
	%   no gravity.
	M = frne(robot, ones(n,1)*q', zeros(n,n), eye(n), [0;0;0]);

	%    compute gravity and coriolis torque
	%    torques resulting from zero acceleration at given velocity &
	%    with gravity acting.
	tau = frne(robot, q', qd', zeros(1,n));	

	qdd = feval(@inv, M) * (torque(:) - tau'); %using builtin function

Loading data, please wait...