Neural code

Question 1

What is neural code?

Answer 1

Information is encoded to communicate & process (we pick up information continuously, information is encoded everywhere – in books, e-mails, conversations)
Neural code (neurons use AP signals to communicate with each other) is at the core of brain function.

Question 2

What are the different spatial principles of brain networks?

Answer 2

Local: small generic processing units (small networks, e.g. three neurons which communicate with each other through synapses).

Global: aggregate related processes in units: dedicated whole brain regions: location code (map).

Large-scale connectivity: intra-cortical pathways/ streams = information highways.

Question 3

What is the basic principle in the visual system?

Answer 3

The basic principle in the visual system is you have a stimulus that is thrown into the eye and it responds by creating neural activity in the optic lobes.

Question 4

What is neural response?

Answer 4

It can be interpreted as the excitation from the soma of a neuron.
(It’s the exponential multiplication of a tiny signal).

Question 5

What is rate code?

Answer 5

Rate code is the the average number of spikes per unit time.

As the intensity of a stimulus increases, the frequency or rate of action potentials, increases.

Information is encoded in the firing rate. There is a coding in the stimulus in terms of the number of spikes per second.

Question 6

What are the first steps of encoding, from light to spikes?

Answer 6

First the stimulus falls on the retina.

Photoreceptors in the retina have graded potentials. Which connect to interneurons through synapses which often operate on graded potentials.

Then they go on to a neuron which ‘counts the number of spikes.’ The spikes represent the intensity of the stimulus. They are carried forward to the next level of information processing.

Question 7

What are the features of a direct one-to-one representation of each sampling point?

Answer 7

The neurons detect the single point and transmits this further into the next level as a single neuronal activity.
There is no interaction between the information coming from different points in the visual field.
• size not encoded (response level stays the same)
• low sensitivity
• high resolution

Question 8

What are the features a direct many-to-one representation of each sampling point?

Answer 8

These neurons have many inputs and create one output.
The neurons interact and converge through summation.
• graded response (response gradually increases for every additional stimuli point)
• sensitivity increased
• reduced resolution

Question 9

What happens after the information processing from the sensory input system?

Answer 9

After information processing from the sensory input systems which is spatial pooling (summation), you can build other elements such as neural computation (e.g. subtraction).

Question 10

What is lateral inhibition?

Answer 10

A neural computation in which responses from neighbouring regions are subtracted is called lateral inhibition.

Question 11

What are the two fundamental types of activity in neural computation?

Answer 11

Two fundamental types of activity: excitation from the summation in the centre which is positive and inhibition from the right side which is subtracting the output from that.

Question 12

What is the first level of processing in the brain?

Answer 12

• Direct representation (one-to-one) of each sampling point (pixel by pixel to the first level of processing in the brain).

Question 13

What is the next level of processing in the brain?

Answer 13

• The next level is summation (Many-to-one)
• This can be a simple summation, and everything is added or it can be a subtraction so it can be processed further.
Involved in lateral inhibition.

Question 14

What is the definition for the visual receptive field?

Answer 14

The location in space where the presence of a visual stimulus can produce a change in the response of a neuron. (This is where the neuron receives information from the outside world)

Question 15

What is the simple mapping of receptive fields?

Answer 15

Shining small spots of light and checking where a change in response occurs.
The number of light spots you detect, gives you an idea about the size of the receptive field.

Question 16

What is a two-dimensional receptive field?

Answer 16

Neural computation in two dimensions: in a receptive field, light is collected from a restricted region of the visual field (a single line of receptors) but they are spread all over in two dimensional images. There can be two types of that:  
Some of them are excitatory surrounded by inhibition.
Or inhibitory (giving negative responses) and being surrounded by excitatory positive responses. (On Off centre)

Question 17

What is an on off centre?

Answer 17

This is a centre where the response of the neuron increases and there is a surrounding area where response decreases when we put a light there.

Question 18

How can you record a two dimensional receptive field?

Answer 18

Plotting receptive fields with a spotlight. E.g. retinal ganglion cells
The stimulus is the monitor and the person looking at it is given point stimuli. Sometimes there is a positive response, no response or a negative response.

There is a centre where the response of the neuron increases and there is a surrounding area where response decreases when we put a light there. This is why it’s called a centre surround organisation.

Alternatively, it can also be recorded by a larger stimuli instead of point by point using an electro physical recording from neurons. (See page 9 of notes for diagram).

Question 19

What is the spatial integration in two dimensional cases?

Answer 19

Adding up excitation and inhibition.

It’s almost like balancing:
The more light in the centre, the more positive the response and the more light into the surround, the more negative the response.

You balance this up to find your final result.

Question 20

What happens when there is a difference between what happens in the centre and surround.

Answer 20

This is in some way a level of intelligent encoding.
By balancing positive and negative responses, you take out the average which isn’t interesting, and you know where the differences/ contrasts are.

Question 21

What is meant by contrast (where the differences are)?

Answer 21

Contrast is the relative difference in the stimulus intensity between a bright and a dark region of a visual stimulus.

Question 22

What happens at a luminance boundary (contrast border)?

Answer 22

• In regions of equal luminance, excitation and inhibition cancel each other.
• At the boundary the excitation and inhibition are not balanced and thus increase the relative difference of perceived brightness.
• At the bright side of the boundary, you have more excitation than inhibition and that’s why you get a positive response.
• On the dark side of the contrast border, you have more inhibition than excitation because the excitatory neuron is not stimulated. So, you get a negative response here and a positive response there.

• Boundary encoding is a very clever way of encoding because it saves us a lot of activity – you only present electric activity in the brain when things change.

Question 23

What is meant by the process of filtering?

Answer 23

Filters are selection mechanisms for information. The filter mechanisms reduce uniform information and focus on the important thing.
• The information is being transformed
• Regions of change are emphasised
• The receptive field acts as a filter – signal from around boundary regions are transmitted, but not from uniform areas.

Question 24

What is meant by redundancy reduction?

Answer 24

We don’t transmit information which doesn’t contain any change. Only when something is changed, you transmit the information, and this is maximising redundancy. E.g. when setting an alarm, it only rings when something happens and if it was ringing all the time it would be useless.

Question 25

What does the receptive field function do?

Answer 25

• contrast enhancemnet
• redundancy reduction
• spatial filtering

Question 26

What is spatial filtering?

Answer 26

The size of a receptive field determines spatial detail visible – different images are visible at different spatial scales (levels of detail).
We are filtering things to give us information for a particular level of detail.

Can be compared with a gravel pit, the brain is like a pit where you store things, you can have a mix of rocks and sand.
We have spatial frequency filters (like the sandpit). We have filters which filter out the large, middle and small parts.
They match the size of the receptive fields.

Question 27

What is spatial frequency?

Answer 27

Spatial frequency is the number of cycles per degree of visual field. A high spatial frequency is where you have lots of stripes, but you have a small number of stripes or cycles in a low spatial frequency.

• (One-dimensional) ‘spatial frequency’ is the number of cycles of a periodic pattern per degree of visual angle

Question 28

When is spatial detail used in images?

Answer 28

Spatial scale is used by artists to draw attention to detail. When you get more and more lines, they blur to one grey or black area depending how dense they are.

• Painters use their skills to present different levels of spatial detail. Spatial detail can be skilfully used as an artistic tool. In some paintings, we can select levels of spatial scale, by blurring the image (thus removing high spatial frequencies) to recognize different things.

Question 29

List some hierarchical encoding stimulus attributes in images.

Answer 29

• Location
• Brightness and colour
• Contrast
• Spatial frequency, detail
• Line orientation, thickness, curvature
• Motion
• Length, size, shape
• Texture, features, objects, faces
• And so on……

Question 30

What is meant when we say that neural encoding is a hierarchical system?

Answer 30

It means we can have a look at encoding at the individual level of individual neurons or individual layers in the visual stream.

We can look at interactions between individual layers. You can see how these interact again with others and others.
You can combine information and create filters for more and more complex information.

In the beginning you just have the encoding of location, you have 1-5-megapixel picture (one-by-one) each has a colour intensity with a location code.

You can then extract colour and brightness and you can look at contrast instead of individual pixels. You can see how pixels interfere with each other.

You can look at spatial frequency and at different detail. The thickness of the line, whether it’s curved. Motion is another thing which is encoded - is it fast or slow or still.

Question 31

What is serial/ parallel processing?

Answer 31

Hierachical integration of neurons generates novel receptive field properties like orientation.

If you align receptor fields (vertically and diagonally) and sum them up. In the first case you would be building a filter for a vertical line and the other would be building a filter for a diagonal line.

It’s a bit like playing with Lego: from circular centre surround receptive fields, we build oriented line segment filter…

All of this is a combination of serial and parallel processing. Very clever system to build an intelligent information processing machinery which is based on encoding information.

Question 32

How do neurons respond preferentially/specifically to local features?

Answer 32

When you look at any object in your surroundings like a flower for example, your brain immediately looks at brightness, size, orientation, colour, texture, emotion and so on.
There are feature specific filters or elements or coding mechanisms which account for all of what we perceive.

They are organised in a parallel way and the receptive field structure is a core element of the whole thinking around that.
This is the area from which you take the information and how this information at particular areas should look like in order to give you a strong response, which identifies that.

Question 33

What is the difference between a high and low level of analysis when looking at a picture/ painting? E.g. the Mona Lisa?

Answer 33

Da Vinky??

Low level analysis: encoding features - brightness, colour, contrast, spatial detail, orientation, texture

High level analysis: understanding the meaning - the famous smile.

Question 34

What is encoding a basis for?

Answer 34

Perception

Question 35

What is a receptive field?

Answer 35

The location in space where the presence of a stimulus can produce a change in the response of a neuron.

Question 36

What is summation?

Answer 36

The neural computation that results in adding together the responses of neurons.

Question 37

What is filtering?

Answer 37

The process in which signals from around boundary regions are transmitted, but not from uniform areas.

Question 38

What does the brain encode?

Answer 38

Identity, intensity/frequencies and tuning.

Question 39

What different terms indicate the neural computation that results in adding together the responses of neurons?

Answer 39

Pooling
Integration
Summation