Cerebellar learning using perturbations
Boris
Barbour1, 2, 3, 4,
Guy
Bouvier1, 2, 3, 4,
Claudia
Clopath5,
Célian
Bimbard1,
Jean-Pierre
Nadal1, 2, 4, 6,
Nicolas
Brunel7 and
Vincent
Hakim1, 2, 4
-
1
Ecole Normale Supérieure, France
-
2
CNRS, France
-
3
INSERM, France
-
4
PSL, France
-
5
Imperial College London, Biotechnology, United Kingdom
-
6
EHESS, France
-
7
Chicago University, Neuroscience and Statistics, United States
The cerebellum aids the learning and execution of fast coordinated movements, with acquired information being stored by plasticity of parallel fibre--Purkinje cell synapses. According to the current consensus, erroneously active parallel fibre synapses are depressed by complex spikes arising as climbing fibres signal movement errors. However, this theory cannot solve the credit assignment problem of using the limited information from a global movement evaluation to optimise behaviour by guiding the plasticity in numerous neurones. We identify the possible implementation of an algorithm solving this problem, whereby spontaneous complex spikes perturb ongoing movements, create an eligibility trace for plasticity and signal resulting error changes to guide plasticity. These error changes are extracted by adaptively cancelling the average error. This framework, stochastic gradient descent with estimated global errors, generates specific predictions for synaptic plasticity rules that contradict the current consensus. However, in vitro plasticity experiments under physiological conditions verified our predictions, highlighting the sensitivity of plasticity studies to unphysiological conditions. Using numerical and analytical approaches we demonstrate the convergence and estimate the capacity of learning in our implementation. Finally, a similar mechanism may operate during optimisation of action sequences by the basal ganglia, where dopamine could both initiate movements and signal rewards, analogously to the dual perturbation and correction role of the climbing fibre outlined here.
Acknowledgements
This work was funded by ANR-08-SYSC-005 and the Labex MEMOLIFE (BB) and NSF IIS-1430296 (NB). GB was funded by Région Ile-de-France, FRM, and Labex MEMOLIFE. This work was supported by the program `Investissements d’Avenir' from the French Government, implemented by Agence Nationale de la Recherche (ANR), references: ANR-10-LABX-54 MEMOLIFE and ANR-11-IDEX-0001-02 PSL* Research University.
Keywords:
Cerebellum,
learning and memory,
synaptic plasticity,
stochastic grdient descent,
credit assignment problem,
reinforcement learning,
Climbing fibres,
Purkinje Cells,
complex spike,
inferior olive
Conference:
The Cerebellum inside out: cells, circuits and functions
, ERICE (Trapani), Italy, 1 Dec - 5 Dec, 2016.
Presentation Type:
poster
Topic:
Cellular & Molecular Neuroscience
Citation:
Barbour
B,
Bouvier
G,
Clopath
C,
Bimbard
C,
Nadal
J,
Brunel
N and
Hakim
V
(2019). Cerebellar learning using perturbations.
Conference Abstract:
The Cerebellum inside out: cells, circuits and functions
.
doi: 10.3389/conf.fncel.2017.37.000023
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
18 Nov 2016;
Published Online:
25 Jan 2019.