A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD - PowerPoint PPT Presentation

Loading...

PPT – A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD PowerPoint presentation | free to view - id: 142a03-MjMwO



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD

Description:

A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD ... Offloading learning from the OV to the FOR resolves this problem. ... – PowerPoint PPT presentation

Number of Views:45
Avg rating:3.0/5.0
Slides: 17
Provided by: krasn
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: A Computer Model of Saccadic Adaptation Reveals the Insufficiency of Cerebellar LTD


1
A Computer Model of Saccadic Adaptation Reveals
the Insufficiency of Cerebellar LTD
  • J.L. Krichmar1, G.A. Ascoli1, L. Hunter1,2 and
    J.L. Olds1,3., 1Krasnow Institute for Advanced
    Study at George Mason University, Fairfax, VA.,
    2National Library of Medicine, Bethesda, MD., 3
    Uniformed Services University of the Health
    Sciences, Department of Anatomy and Cell Biology,
    Bethesda, MD.

2
Abstract. The oculomotor vermis region (OV) of
the cerebellar cortex and the fastigial ocular
region (FOR) of the deep cerebellar nuclei
appears to be necessary for the execution and
maintenance of accurate saccades. A recent
computer model of a Purkinje cell (PC) population
demonstrated the ability to qualitatively
reproduce OV PC recordings during saccades. This
model has detailed dendritic spines with a
biologically plausible cerebellar LTD learning
rule. In rapid adaptation gain increase
simulations, our model was able to adjust its
gain in response to hypometric saccades to a test
target. This had the expected result of gain
transfer to adjacent directions (?30o) and
unaffected post-adaptation saccades to
contralateral targets. In rapid gain decrease
simulations, our model adjusted its gain
appropriately. However, simulations of post
adaptation saccades were dysmetric on saccades to
targets both ipsilateral and contralateral to
adaptation. We incorporated a membrane
potentiation learning (LTP) rule into our model
to complement LTD. LTP resolved the dysmetria of
the gain reduction simulations, but the
responsiveness of our modeled PCs to inputs
became limited after adaptation. Adding the
ability of the simulated PCs to entrain cells in
the FOR resolved this problem. It appears from
our saccadic adaptation experiments that 1) PCs
need to both increase and decrease their
excitability to adapt to change. 2) Plasticity
in the cerebellar cortex alone is not sufficient
for a complete range of gain change. This
research was supported by the Krasnow Institute
for Advanced Study.
3
Introduction
  • A novel computer model of cerebellar saccadic
    control has elucidated
  • learning a motor pattern requires both
    potentiation (LTP) and depression (LTD) learning
    rules.
  • the location of the acquisition of motor learning
    patterns is in the cerebellar cortex (CC).
  • the location of the long-term storage of a motor
    learning pattern is in the deep cerebellar
    nucleus (DCN).

4
Saccadic Control
Higher Cortex, Visual System and Pons
Eye Position
Target Position
Saccadic Cerebellum
Climbing Fibers
Motor Efference Copy
Proprioceptive Information
Fine Motor Command
Coarse Motor Command
Superior Colliculus
Brainstem Saccade Generator
Inferior Olive
NRTP
Motor Efference Copy
Oculomotor Command
5
Cerebellar Saccadic Control
ipsilateral
contralateral
eye position
Oculomotor Vermis
muscle proprioceptors motor efference copy
before and during saccade
end of saccade
-
-
Fastigial Ocular Region
end of saccade
before and during saccade


Brainstem Saccade Generator
6
Model of Cerebellar Saccadic Control
  • A network model of the oculomotor vermis (OV).
  • The model contains 256 Purkinje cells.
  • 295,424 total compartments
  • Each Purkinje cell contains
  • 1024 dendritic spines - receiving a parallel
    fiber input.
  • 128 dendrites - receiving a stellate cell and
    climbing fiber input.
  • 1 soma - receiving a basket cell and climbing
    fiber input.
  • Each Purkinje cell is initialized to
  • Fire minimally at a preferred direction.
  • Fire early in a saccade (early burster) or late
    in a saccade (late burster).
  • The output of the OV is a vector summation of all
    the Purkinje cells firing rates.
  • A black box model of the Fastigial Ocular
    Region (FOR) with gains for every direction.
  • Test model with rapid adaptation paradigm.

7
Methods Qualitative Reasoning Neuron (QRN)
  • Vocabulary of QRN
  • Landmarks - describe the critical values for a
    parameter
  • Parameters
  • Continuous
  • Discrete
  • Quantity Space - set of landmarks
  • Qualitative State - the state of a parameter
  • Direction - increasing, decreasing, steady
  • Value - at or between landmarks
  • Weight - relative magnitude of a parameter
  • Constraints
  • M(a,b) Monotonically increasing
  • M-(a,b) Monotonically decreasing
  • EXPD(a) Exponential decay
  • EXPI(a) Exponential increase
  • TH(a,b) Threshold step function.
  • (a,b,c) Addition
  • (a,b,c) Multiplication

8
(No Transcript)
9
Learning Rules
10
  • Adaptation occurs after each movement
  • Parallel fibers contain error signal with
    direction and magnitude.
  • Climbing fibers contain a signal that denotes an
    evaluation phase (all PCs receive the same CF
    input).
  • Only adjust the early bursting Purkinje cells.
  • Gain increase is achieved through post-synaptic
    LTD of PF-PC synapse1.
  • Gain decrease is achieved through pre-synaptic
    LTP of PF-PC synapse2.
  • After adaptation, learning is transferred from
    the OV to the FOR by burst type and direction3.

1LindenConnor (1995), Ann. Rev. Neur., 18,
318-57. 2Salin et.al (1996), Neuron, 16, 797-803.
3Mauk (1997), Neuron, 18, 343-346.
11
Results of LTD Only Adaptation Experiments
  • Rapid adaptation to 30 overshoot of horizontal
    left target.
  • Model correctly decreases gain on movements to
    the left
  • Model incorrectly increases gain on movements to
    the right.

12
  • Firing rates of Purkinje cell (PC) population in
    the simulated OV.
  • Each solid line represents the firing rate of a
    single Purkinje cell.
  • Direction of line denotes PC preferred direction.
  • Length and color denotes PC firing rate.
  • Dotted blue line denotes vector summation of OV.

13
Results of LTD and LTP Adaptation Experiments
  • Matches psychophysical data.
  • Gain changes are transferred only to leftward
    movements.
  • Increased inhibitory output from the Purkinje
    cell population decreases the gain (see figure on
    right).

14
Results of Adaptation and Recovery Experiments
15
  • Pre-synaptic LTP and post-synaptic LTD of the
    same PF-PC synapse causes desensitization of the
    PC responsiveness.
  • Offloading learning from the OV to the FOR
    resolves this problem.
  • However, some saturation still occurs as the
    magnitude FOR becomes larger with respect to the
    OV.

16
Conclusions
  • Both LTP and LTD in the CC are necessary for
    motor learning.
  • Our simulations suggest
  • The DCN is where motor patterns are stored.
  • Storage at the DCN allows for large range of
    responsiveness in the CC.
  • This model of saccadic eye movements can
    generalize to other movements.
About PowerShow.com