Cerebellar Model

Question 1

What are the two possible sites of plasticity?

Answer 1

Cerebellar cortex
Deep cerebellar nuclei

Question 2

What is the Marr-Albus Framework?

Answer 2

To learn to make accurate movements, you must have information about what you did wrong- an ‘error signal’

Climbing fibres convey that error signal

Question 3

What are the models trying to do?

Answer 3

Explain how the cerebellum is important for motor control

Question 4

How can we start trying to understand what motor control invovles?

Answer 4

Through task analysis

Question 5

What is the most abstract level that complex information processing can be modelled at?

Answer 5

Computational level
-understanding what is the cerebellum trying to solve

Question 6

If we want to look at a target that is to the right, what do we need to know?

Answer 6

The size of the command we need to send to the muscles

Eventually need to know how big a command to send to the muscles wherever the target is located

Question 7

To figure out HOW the cerebellum works out the size of the command what level do we need to examine?

Answer 7

Algorithmic level

Question 8

How are precise motor commands learnt for the cerebellum?

Answer 8

Supervised learning

Question 9

What is supervised learning?

Answer 9

If you make an error THEN you correct what you do - error signal indicates which direction to alter the command

So the error signals can be considered a form of supervision

Question 10

How can the correct value of a motor command be learnt?

Answer 10

Using an error signal

Question 11

What is the learning rule?

Answer 11

If a movement is too small, increase the command

If a movement is too big, reduce the commans

If accurate, do nothing

Question 12

What is implementation?

Answer 12

How the cerebellum can carry out the corrections to movement command size

Question 13

How is output (how much the PC fires) determined?

Answer 13

Depends on the weight of the synpase between the parallel fibres and the purkinje cell
- that is the weight that is adjusted by the climbing fibre error signal

if movement too small - increase weight
if movement too big - decrease weight

Question 14

What is the idea of the algorithm then?

Answer 14

You eventually learn how big a command to send to the muscles wherever a target is located
- this is true of any robot or device making accurate movements

Question 15

What can the method that learns to approximate the function that relates motor command to retinal position be called ?

Answer 15

Function approximation method

Question 16

Is the idea of supervised learning plausible?

Answer 16

Yes

Explains generally why there may be so many granule cells (80% of all neurons)

A lot of factors can affect the size needed for an accurate motor command and that perhaps can be used to illsutrate the need for constant practice e.g. in music or sport
- a lot of information is needed to get the movement just right

Question 17

What is another reason that supervised learning is plausible?

Answer 17

Explains why CF input is so unsual
- its low frequency (1 spike/second) means that error signal does not drive output (we dont want it to drive output)

Very relaible- whenever CF fires, PC also fires and the whole dendritic tree is effected, all 150,000 synapses between granule cells and PCs affected
- just what is wanted from an error signal

Question 18

What is the decorrelation learning rule?

Answer 18

Synaptic weight is changed according to the correlation between the parallel-fibres signal and the error signal conveyed by the climbing fibre
positive correlation = reduce the weight
negative correlation = increase the weight

Learning stops when there is no longer a correlation between any parallel-fibre signal and the climbing fibre signal

Can therefore be called a decorrelation learning rule

Question 19

How does this apply to eyeblink conditioning?

Answer 19

If the initial movement is too small e.g. no blink at all

The error signal (UCS) needs to increase movement size

Becuase of the circuitry, this measn the UCS should teach purkinje cells to fire LESS

Question 20

What happens before conditioning?

Answer 20

Error signal =UCS (air puff to eye), sent by inferior olive

CS =tone, via the mossy fibres
CR = produced by change in Purkinje cell firing

Before conditioning the CS produces no response, i.e. no change in Purkinje cell firing

Question 21

What happens during conditioning?

Answer 21

Mossy (Parallel) fibre CS is now paired with climbing-fibre UCS

This pairing produces LTD in the excitatory synapses between parallel fibres and Purkinje cells

Makes functional sense – CS predicts being blown in the eye, i.e. an ERROR! Do this less

This is a synaptic process, which the model can connect to function

Question 22

What is then the predicted site of plasticity?

Answer 22

Synapses between granule cells and purkinje cells

Question 23

What still remains unknown?

Answer 23

Whether the mechanism proposed by the model is correct
- its plausible but does it fit with the data?