6- Neural Correlates of Perceptual Learning Flashcards

1
Q

Representation and readout of visual signals:

We have a hierarchy of visual processing starting from the retina to the dorsal and ventral visual streams. Along this pathway visual input becomes more…?

A

complex

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

To make a motor or perceptual decision, the representation of visual input in visual cortex has to be

A

“read out” or decoded

-V5 area identifies, but higher order processing takes place via other brain areas to make the decision based on different sensory information

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Which 3 brain areas decode neural signals in the visual cortex to make motor or perceptual decisions?

A

Lateral intraparietal (LIP),

Frontal eye fields (FEF),

Prefrontal (area 46) cortex

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

What determines the output of the model/behaviour of the organism in terms of Typical neural model of visual discrimination to make a response?

A

The population of neurons projects onto a single output (info from all neurons are pulled together from different outlets in order to be transferred into making a decision)

The 2 ways we can do this is by:

1- weights (pull info from each unit and add it together to make decision)
2- based on specific demands from the task (apply different weights to each unit to get different combinations of weights in different units)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Typical neural model of visual discrimination:

Describe this model

A

At the level of the visual representation, a population of neurons encode the visual input (e.g. direction of motion, 45* orientation)

Each neuron responds to specific info represented in a specific way

Each neuron is subject to internal noise (same stimulus presented to the same neuron 10 times, the response they provide will be slightly different each time). Responses are variable, even to repeated presentations of the same stimulus

The population of neurons projects onto a single output, whose value determines the output of the model/behaviour of the organism

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Decoding neural signals to form perceptual decisions:

How is the visual input (e.g. direction of motion) is represented or encoded?

info slide

A

by the pattern of firing across a population of neurons

Perceptual decision: Is the direction A or B?

Direction A and Direction B generate a different pattern of responses from the population of neurons (some neurons will respond exactly to the direction giving a maximum response but some will respond a little to the surrounding which still gives some information on the direction of motion)

This obtains 2 population responses:
-which allows us to determine the discrimination boundary (minimum overlap between 2 discrimination responses)

To make an actual decision:
-Readout assigns negative weights to neurons that encode A and positive weights to neurons that encode B

This generates a decision variable whose sign enables a decision making brain area (e.g. LIP) to distinguish between two alternatives (choose Direction A or Direction B)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Specificity of learning for low-level visual characteristics (e.g. orientation, retinal position) led to the view that neurons in early visual cortex are malleable or ‘plastic’.

But specificity of perceptual learning does not necessarily indicate that learning is changing how learned stimuli are represented in visual cortex. Specificity could equally indicate that …

A

learning is changing how visual information is read-out to form decisions (might be that we are becoming better at pulling information together)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

How could perceptual learning change the representation in visual cortex:

Perceptual learning could change the responses of individual neurons

Learning could change the preferences of neurons, width of tuning functions (and slope) or amplitude of response

These changes to individual neurons could alter the representation of stimuli in visual cortex.

A

improvements may manifest itself via
-bandwidth (how tuned is that orientation, the wider the distribution the less tuned it is to firing to a specific characteristic but the narrower the bandwidth the more specific it is to firing), tuning

-amplitude (how much does a neuron spike when a specific stimulus is presented)

increase in these after training

-baseline firing= amount of noise (neurons are consistently firing at a baseline level, but to encode info it has to pass a spiking threshold to recognise that that info is present, the baseline decreases over trials to encode info)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Perceptual learning changes tuning of neurons in primary visual cortex?

Monkeys

A

Monkeys learned to improve their orientation discrimination performance at one (trained) location in the visual field

The learned improvement correlated with a very subtle change in the slope of orientation tuning curves at the trained location

Study shows that there is some change in the visual representation of the orientation but it isn’t from the cells which encode but the ones that are slightly off (off axis neurons)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Perceptual learning could change the
pattern of correlated neuronal responses:

Responses of neurons with similar … are more strongly correlated than neurons with …?

A

tuning preferences, dissimilar tuning

Learning could change how neuronal populations are correlated with each other

(with learning the level of noise is lowered, to make it easier for the units to provide responses)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

With training, something at the level of representation is changing, what does this mean?

A

the ones that are slightly off (off axis neurons)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

True of false:
Neither V1 or V2 shows changes in representation in visual information

A

True

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Perceptual learning could change the
pattern of correlated neuronal responses:

The responses of neurons in visual cortex are weakly correlated with each other, which can have substantial effects on …?

A

the information encoded by neurons

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

How could perceptual learning change the read-out?

Learning could optimise the read-out for current perceptual decision by giving more weight to the neurons that are providing…?

A

the most reliable information

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Area … represents motion information that is used to solve the task

Area … represents the transformation of motion information into an eye movement (saccadic) choice and reads out info

A

MT , LIP

results:
In monkeys
MT neurons responded similarly to motion throughout the training period

LIP neurons were insensitive to visual motion at the beginning of training, but their responses grew in strength as performance on the task improved during training

Strongest evidence that perceptual learning changes the read-out but not the representation of visual information

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Does perceptual learning change the neuronal “representation” of visual stimuli in cortical visual hierarchy?

A

Learning does change the responses, tuning and correlated responses of neurons at intermediate levels of the cortical visual hierarchy

BUT:

These changes to the neuronal representation of visual stimuli in visual cortex are probably not sufficient to account for the behavioural changes brought about by perceptual learning

6
Q

The strongest evidence is that perceptual learning changes the … but not the representation of visual information?

A

read-out

improvement happens at the level of LIP, there is no change in MT

7
Q

How does learning on a new task affect the neural substrates of previously learned tasks?

Examined the effect of fine discrimination training on the causal contribution of MT to coarse depth discrimination. What 3 different tasks were used?

A

Coarse depth discrimination task

Fine depth discrimination task

Direction discrimination task

Trained monkeys to perform the :
Coarse depth discrimination
Direction discrimination

=Reversible inactivation of MT devastated monkeys’ performance on coarse depth discrimination and direction discrimination

8
Q

Which task is this?

Signal dots presented either near or far (noise dots at random depths). Monkeys reported whether net depth of dots was near or far

A

Coarse depth discrimination

8
Q

Which task is this?

Surround dots at fixed depth and centre dots presented at different depths. Monkeys reported whether centre dots were in front or behind surround dots

A

Fine depth discrimination

8
Q

It is possible to inactivate MT (which is still reversable) when performing which 2 tasks?

A

Coarse depth discrimination
Direction discrimination

8
Q

Which task is this?

Signal dots moved either right or leftward (noise dots in random directions). Monkeys reported the direction (left or right) of global motion

A

Direction discrimination

8
Q

Learning can change how visual information is read out from visual cortex to form …?

9
Q

How does learning on a new task affect the neural substrates of previously learned tasks?

Area MT is not involved in depth/fine discrimination tasks

A

Reversible inactivation of MT devastated monkeys’ performance on coarse depth discrimination and direction discrimination

BUT:

After same monkeys trained on fine depth discrimination task , muscimol injection into MT produced no effect on coarse or fine depth discrimination, but abolished direction discrimination performance

Strongly suggests training on one task changes the causal contribution of a brain area to another (related) task.
This effect did not change disparity tuning in MT, again suggesting plasticity in the read-out

10
Different types of visual information are represented at different levels of the visual cortical ...?
hierarchy
10
To make a visually guided decision, the representation of visual information in visual cortex must be ... by decision making networks
“read out” or decoded Just having a visual representation is not enough These substantive alterations to the read-out are sufficient to explain changes to the perception of visual motion
11
True of False? Learning on one task can't change the causal contribution of a brain area to another similar task without changing the representation of visual information
False it can