Brainscape's Knowledge GenomeTM

Browse over 1 million classes created by top students, professors, publishers, and experts.


See full index

Brain and Behaviour > Principles of Neural Representaion > Flashcards

Principles of Neural Representaion Flashcards

You may prefer our related Brainscape-certified flashcards:
The Brain flashcards
1
Q

How do neuroscientists define representation?

A

The way different kinds ofinformation are transmitted in the activity of neurons or groups of neurons

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

How does Vilarroyal 2017 define representation?

A

A neural representation is a pattern of neural activity that stands for some environmental feature in the internal workings of the brain

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

What is processing?

A

Systematically transforming one pattern of activity into another

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

How is a neuron compared to a detector?

A

It integrates information from a variety of sources, and sends a signal reflecting the degree to which the inputs match some pattern

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

What do extracellular single unit recordings measure?

A

Electrical activity generated by individual neurons near the tip of a metal electrode

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

How do you interpret a raster plot?

A

Each dot represents a spike

Each row is an experimental trial

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What is a head-direction cell?

A
  • Found in the post-speculum of the rat

- Represents the direction that the rats head is facing

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What are the limits of the detector analogy?

A
  • Does the detector idea require a unique neuron for every conceivable percept, concept, or action?
  • Ambiguity in the firing rate of an individual neuron
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

What proportion of stimuli given to a MTL neuron will make it fire?

A

0.54% of possible stimuli

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What is a MTL neuron?

A

Medial Temporal Lobe Neuron

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What does it mean if a neuron response is highly invariant?

A

The responses are not sensitive to superficial variations (pose, lighting, etc)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Why are broad tunings useful?

A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Why are broad tunings useful?

A
  • They mean that the cell can communicate some useful info, even when the conditions it is detecting are not met precisely
  • Make it possible to generalise
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

The firing rate of a neuron can…?

A
  • Signal a degree of match between the current input and its ‘preferred value’
  • Be interpreted as the probability the particular stimulus is present, the degree to which something appears to be true, or some action is intended
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

How does considering an additional neuron resolve some of the ambiguity?

A

The neurons will have different preferred values and so different firing patterns and so we can distinguish between those different stimulus values

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Why can similar patterns be useful?

A

Processed in a similar way - generalisation

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Why can similar patterns be problematic?

A

May be difficult to distinguish patterns that need different processing

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

What is coarse coding?

A

Each neuron represents a range of the possible input values

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

How are accurate estimates of coded information reconstructed?

A

By pooling info from different neurons, e.g. by taking a firing weighted average of the preferred values of each neuron

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

How do we show that we understand a code?

A

By decoding it

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
20
Q

How do we show that we understand a code?

A

By decoding it

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
21
Q

What is the first neural system that encodes colour?

A

Retina

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
22
Q

How are non-spectral colours represented?

A

Combinations - represented as distinct patterns

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
23
Q

What physically possible colours can’t be distinguished in this code?

A

Metamers - mixtures that produce the same activity can’t be distinguished from pure spectral colours e.g. red+green=yellow

24
Q

What is sparsity?

A

The proportion of neurons that fire in a given time window

- Also related to the proportion of stimuli to which a given neuron responds

25
Q

Why aren’t grandmother cells good for generalisation?

A

Because they could not capture the similarity of different stimuli

26
Q

What are advantages of sparse population codes?

A
  • Highly selective
  • Emphasises uniqueness
  • Resistant to interference
  • Efficient in terms of energy use
  • Allows for rapid learning
  • Allows for simple network architecture
27
Q

What are advantages of dense population coding?

A
  • Broad tuning
  • Emphasises generalisation
  • Captures similarity
  • Redundant
  • Efficient in terms of neurons required
28
Q

What are disadvantages of dense populatio coding/

A
  • May require complex learning rules
  • Prone to interference
  • May require complex architecture
29
Q

Can information from an individual neuron be combined to predict the actual movement?

A

Single neuron activity increases when the lever is about to be moved in the preferred direction. Decreases as the level is moved in the opposite direction

30
Q

What is the neural voting system?

A

Each neuron votes for its prefered direction. The oveall result can be calculated by taking the vector sum of the preferred directions of each neuron weighted by its firing rate

31
Q

Where do face cell population codes take place?

A

Where do face cell population codes take place?

32
Q

Where do place cell population codes take place?

A

Hippocampus

33
Q

Give examples of population codes.

A
  • V4 code for 2D shapes
  • Motor cortical code for limb movement
  • Face cells
  • Place cells
  • Head-direction cells
34
Q

What is neural computation?

A

Processing

  • Systematically transforming one pattern of activity into another
35
Q

What is the purpose of neural network models?

A

To allow us to begin to understand the mechanisms of cognition and behvaiour, while incorporating constraints from neuroscience

36
Q

How does neural computation work?

A

A specific pattern of activiy in one population of neurons induces a specific pattern of activation in another

37
Q

A specific pattern of activity in one population of neurons induces a specific pattern of activation in another

A
  • Each neuron modelled as a uni with no physical stucture except some incoming and outgoing connections
  • Each unit has a property called activation analogous to its firing rate
  • Each connection has a property called its weight which is proportionate to the total strength of all inhibitory and excitatory connections between the neurons
38
Q

How can we work out the activation of a given neuron?

A

Take the sum of its inputs weighed by these connection strenghts, and it’s a function of those that met inputs which gives us the activation of a particular neuron exchange

39
Q

Explain the feed-forward structure

A

Layers of neurons where each unit in one layer is connected to all the units in the next layer

40
Q

Explain supervised learning.

A

Network learns by being presented with example input pattern, and the corresponding desired output pattern.
- Initially output is different to desired because the weights are random but they gradually and systematically change to reduce the discrepancy.

41
Q

What is unsupervised learning?

A

Works by discovering inherent structure in the input.

  • Connections strengthened between neurons that fire concurrently
  • More biologically plausable
42
Q

What is competitive learning?

A
  • Type of unsupervised learning
  • learns to identify clusters in the input patterns
  • the output neurons compete to respond to each input pattern
  • eventually each cluster will be represented by a different output neuron
43
Q

How do we make sure each output unit has a fair chance of responding to each input pattern?

A

The total weight of connections to each output is limited is fixed.
If some connections are strengthened, others must be weakened

44
Q

Briefly, what is supervised learning?

A

The output pattern is compared with the ‘correct answer’ and weights adjusted to reduce discrepancy in the future

45
Q

Briefly, what is unsupervised learning?

A

No correct answer is needed
Competitive learning: output units compete, the weights to the winning are adjusted to that it will be more likely to win in future

46
Q

What is deep learning?

A
  • Multilayer neural networks
  • Using variants of supervised learning
  • Much more data, faster computers
  • Can approach of exceed human performance on a wide range of tasks
47
Q

What are large language models?

A
  • Trained to complete a piece of text by repeatedly predicting the next word
  • Huge datasets from internet
  • Long training
  • Many billions of weights
48
Q

What are neurons tuned to?

A
  • Retinal location
  • Orientation of stimulus
  • Ocular dominance
  • Direction of motion
  • Colour
49
Q

How do Hubel and Wiesel suggest orientation tuning varies?

A

Orthogonally with ocular dominance stripes

50
Q

How does Bower suggest orientation tuning is organised?

A

Around pinwheels which are evenly spaced along the stripes.

51
Q

What is retinotopy?

A

Systematic mapping between areas of the visual field and areas of the cortex or other brain regions

52
Q

What is tonotopy?

A

The spatial arrangement of where sound is perceived, transmitted, or received

53
Q

What is somatopy?

A

The point-for-point correspondence of an area of the body to a specific point on the central nervous system - typically the area of the body corresponds to a point on the primary somatosensory cortex (postcentral gyrus)

54
Q

Are maps only restricted to primary cortices?

A

No

  • As we get further from primary sensory and motor cortex maps are harder to find because:
  • neurons no longer respond to simple sensory/motor variables
  • we often do not know what the key variables are
55
Q

What is cortical magnification?

A

More sophisticated processing is always reflected in a greater allocation of cortical space

56
Q

What is somatotopic cortical magnification?

A

In somatosensory cortex body parts which are more densely arrayed with receptors are accorded more space in the cortical map than less sensitive regions.

57
Q

What is a centre-on surround-off organisation?

A

Physically nearby neurons facilitate one another, while distant units inhibit one another

Brain and Behaviour (5 decks)

Brainscape helps you reach your goals faster, through stronger study habits.
© 2024 Bold Learning Solutions. Terms and Conditions