Cerebellar Model

Question 1

What is the main theory of motor learning?

Answer 1

Marr-Albus Framework

Question 2

What does the Marr-Albus framework try to do?

Answer 2

Explain how the cerebellum is important for motor control

Question 3

What are the levels in the Marr-Albus framework?

Answer 3

1. Computational level (task analysis)
2. Algorithmic level (method)
3. Implementational level

Question 4

What is the computational level?

Answer 4

Complex information processing problems need to be modelled at more than one level - the most abstract is the computational level

–> What problem is the cerebellum trying to solve?

Framework associated with David Marr

Question 5

What is associated with the computational level?

Answer 5

Task analysis

Question 6

Explain task analysis

Answer 6

Start with a simple task - target shown to a pp
Target appears on the retina 10 degrees to the right
Natural thing is to move the retina to look at the target
To do this we need to know how big a command to send to the muscles
Eventually, we need to know how big a command to send to the muscles wherever the target is located

Question 7

How are motor commands learnt?

Answer 7

Algorithmic level
For the cerebellum - precise motor commands are learnt using supervised learning

Question 8

What is supervised learning?

Answer 8

Supervised learning is characterized by teaching signals that can report whether or not expectations match outcomes (i.e. a yes or no signal)

Question 9

Is supervised learning a plausible explanation?

Answer 9

Yes - as adults we make accurate movements ‘naturally’
But babies spend an enormous amount of time learning about movements

Question 10

How can the correct value of motor command be learnt?

Answer 10

Use an error signal

Question 11

What is the learning rule?

Answer 11

If movement is too small, increase the command
If too big, reduce the command
If accurate, do nothing

Question 12

How can the cerebellum carry out the learning rule?

Answer 12

1. Parallel fibres signal the location of the target
2. When a parallel fibre fires, so does its target Purkinje cell (the motor command)
3. How much the Purkinje cell fires depends on the weight of the synapse between the parallal fibre and the Purkinje cell
4. Weight of synapse adjusted by climbing fibre signal
5. Increases weight if movement too small, decreases if movement too big
6. Eventually learn how big a command to send to the muscles wherever the target is located

Question 13

What is the function approximation method?

Answer 13

Function approximation helps finding the value of an action when similar circumstances occur

This method learns to approximate the function that relates motor command to retinal position

Question 14

What is the function approximation method an example of?

Answer 14

Supervised learning - the error signal indicates which direction to alter the command

Question 15

Is this a plausible method (function approximation)?

Answer 15

Explains why there might be so many granule cells
A lot of factors can affect the size needed for an accurate motor command - need for constant practice
So a lot of information is needed to get the movement just right

Also explains why the climbing fibre input is so unusual - low frequency (error signals should not drive output), very reliable, all synapses between granule cells and Purkinje cells affected - just what is wanted from an error signal

Question 16

What is not explained using the function approximation method?

Answer 16

No mention of what granule cells actually do to mossy fibre inputs - ‘expansion recoding’ not explained

Learning rule is in fact slightly more complicated
- If PF signal positively correlated with CF signal, decrease weight
- If PF signal negatively correlated with CF signal, increase weight

Question 17

What is the decorrelation rule?

Answer 17

Learning ceases when no correlation between any PF input and the CF signal

Also known as Least Mean Square, delta learning rule

Question 18

How does the learning rule apply to NMR conditioning?

Answer 18

The initial movement is too small - in fact no blink at all
The error signal (US) therefore needs to increase movement size
Because of the circuitry, this means the US should teach Purkinje cells to fire LESS

Question 19

Describe the cerebellum and NMR conditioned reflex pathway

Answer 19

Information about the conditioned stimulus arrives at the cerebellum via mossy fibres
Information about the unconditioned stimulus arrives at the cerebellum via climbing fibres

Light/sound –> pontine nuclei –> mossy fibres –> cerebellum

Error signal = sent to inferior olive –> climbing fibres –> cerebellum

Cerebellum –> Cerebellar nucleus efferents –> Red nucleus

Question 20

Explain the cerebellum before conditioning

Answer 20

Error signal is the US i.e., air puff to eye, sent by the inferior olive
CS is the tone, via the mossy fibres
CR is produced by a change in Purkinje cell firing

CS produces no change in firing rate of the Purkinje cells
CS produces no change in cerebellar nuclei firing rate so no conditioned response

Question 21

Explain the cerebellum during conditioning

Answer 21

CS tone decreases PC firing rate and increases the cerebellar nuclei firing rate creating the conditioned response

Parallel fibre CS is now paired with climbing fibre US

This pairing produces long term depression (LTD) in the excitatory synapses between parallel fibres and Purkinje cells

Makes functional sense - CS predicts being hit in the eye i.e., ERROR - do it less
This is a synaptic process, which the model can connects to function

Question 22

Following the Marr-Albus framework, what is the predicted site of plasticity?

Answer 22

Synapses between granule cells and Purkinje cells

Question 23

How does the cerebellar chip apply to NMR conditioning?

Answer 23

Same basic circuit used for eyeblink conditioning as for movements in general
Thus, a cerebellar model used to explain role in motor control should also apply to classical conditioning