Lecture 3 Flashcards
(18 cards)
1
Q
What are the types of prosopagnosia?
A
- Congenital prosopagnosia: life-long disorder from birth
- Acquired: insult to brain in premorbidly typical individuals (trauma)
- Acquired is more researched
- Similar clinical presentation but differ in key characteristics
2
Q
What is the brain damage in acquired prosopagnosia?
A
- Considerable heterogeneity
- Damage to their core system: right fusiform area
- Cases where damage is anterior
- Damage almost always extends far beyond face selective areas
- Bilateral damage is often associated with more severe prosopagnosia
- Broad patterns suggest differentiation between posterior damage and anterior damage
3
Q
What are face representations like in acquired prosopagnosia?
A
- Considerable heterogeneity in behavioural manifestation = hard to draw conclusions as to what goes wrong
- Apperceptive variant: difficulty in generating a rich enough face representation for discrimination
- Associative variant: difficult in matching visual input to memory representation e.g see the face but cannot get the name or identity
- Rare condition
4
Q
What is the data supporting Apperceptive acquired prosopagnosia?
A
- Task where showed ppts 3 different images, 2 were same, 1 was different (changed feature in image)
- Asked ppts to pick odd one out: healthy controls pick the right image, in patients, there is a lot of variability
- Patients were grouped into what kind of brain damage they had: some patients with injuries in the temporal area performed well above chance but not as well as controls = patients with associative variant
- Patients with occipital injuries perform at chance = apperceptive prosopagnosia
- Damage to core region including FFA leads to a difficulty in encoding configurational information
5
Q
What was an experiment for associative prosopagnosia?
A
- Ask ppts to imagine two people and ask who has the narrower face - questions often are shape or feature related
- Patient that performed in previous task now performs at chance due to identity component
- Damage to anterior regions bilaterally leads to difficulty in matching input to memory representations
6
Q
What is prosopamnesia?
A
- Patient CT fractured their skull in 1982 followed by number of brain surgeries
- Asked patient and control to match sequential face: task requires to be able to generate rich face representations but do not have to recognise them
- Ask patient to identify famous individuals, distinct difference between recognising individuals patient was familiar with (famous prior to brain trauma) and ‘unfamiliar’
- These patients have difficulty in learning new face memory representations - but no other issues with their memory
7
Q
Is prosopagnosia face specific?
A
- Patients have deficits in other domains of object perception
- Large lesions that go beyond face-specific areas = deficits in object perception might be due to damage in non-face areas
- Few cases of pure prosopagnosia suggests it is face-specific
- Extensive assessment suggests highly specific difficulty of generating face representations
8
Q
Are humans face experts?
A
- Typical humans are proficient at tasks involving faces
- Difference between familiar and unfamiliar faces
- British and Dutch observers
- Photos of 2 Dutch celebs taken under natural conditions
- Ppts have to sort the images by identities
- Dutch observers = median: 2 perceived identities, Range: 2-5
- British observers = median: 7.5, range: 3-16
- Typical sorting behaviour was seen = sort same identity into different piles, tendency not to mix identities (good at telling people apart, not good at telling people together)
- Under natural conditions, unfamiliar faces are difficult to identify from photographs
- People’s faces vary differently so brains have to learn the variability of each facial identity (lighting etc) and then discount it
9
Q
What is the application to witness testimony, passport control and forensic settings?
A
- Widespread requirement to use photos to prove our identity
STUDY: - Realistic setup assessed police officers and had to match life target to photograph
- 14% of fraudulent photographs were wrongly accepted as valid - authors did not choose difficult images - therefore underestimation of problem with normal setup
- Ability to correctly match person to photo, if you link that to the experience (0-20y), there is no relationship = anyone can be fooled
10
Q
What specific aspects of faces are important in helping us recognise facial identity?
A
- Receptive fields and orientation tuning in v1, neurones tuned to different orientations
- Neurones are all part of one orientation channel
- Orientation information is important to recognising different identities
- If you filter the neurons in a horizontal way, you can still see/recognise them, but not vertically = no identity
- Can assess how much information is carried with each orientation = tends to be more in the horizontal
11
Q
What are biological bar codes?
A
- Horizontal information is like a bar code specific to each individual = allows to recognise identity
- When you apply different image transformation = should be hard to recognise identity, e.g invert polarity = creates the exact opposite barcode = hard to recognise image
- Can also destroy the original bar code is by inverting the image = in humans, it is harder to recognise faces
- Image transformation change aspects in the image but retain the stripes in the horizontal parts of the image = relationship between stripes are the same via compression and variable pose
12
Q
What was a task looking at inversion and familiarity?
A
- Task with personally familiar faces, showed a target image that might be filtered in different ways: either horizontal/vertical image present and had to match the identity as target face
- Do a fourier analysis where either vertical/horizontal is retained and either shown right side up or inverted
- Ppts are worse independently on how it is filtered if picture is inverted
- In upright condition, ppt much better at task where horizontal image was retained, advantage we have of retaining horizontal information plays a role for upright, not inverted
- Familiarity makes the task easier independently of filtered or upright, the difference between the advantage for horizontal information is not as big for unfamiliar than familiar
- Focus on horizontal stays upright not inverted = use to recognise familiar faces
13
Q
What is face space?
A
- Multi-dimension similarity space e.g distance between eyes etc.
- Faces are represented by location
- Location of a face is determined by values along dimensions of face space
- Faces that are close in face space are perceived as more similar
14
Q
What are the models of faces?
A
- Prototype model/norm based model: average face of all faces seen throughout life, and other faces are in reference to norm - specific faces are determined of angle of arrow from centre
- How distinct a face is how far away it is from the norm
- Exemplar-based model: no such thing as a norm and faces are represented by their absolute position to their multidimensional space - not in relation to norm (no norm, just where they sit) - distinction is how many faces are near them on the dimensions
- Models of how values along dimensions are represented
15
Q
What was evidence for prototypic models?
A
- Identification of caricatures
- Had to identify an image of Daniel Craig - most people say the rightmost image is him, but the middle one is actually him. The right is a caricature, and the left is anti, done by taking a norm male and exaggerating features of male face
- Occurs because you can travel along one direction in the multidimensional model to get the same identity but exaggerated
16
Q
What is opponent vs multi-channel coding?
A
- Associated with ways of coding facial information
- Opponent: in respect to a specific feature in face space, have two types of neuron populations, both neuron will react equally to a feature on average, some neurons will respond to values larger to norm, and second population react to lower than norm = prototypical based model
- Multichannel: Have channel that responds to specific features = low level features = gradient response of optimum neuron = exemplar based model
17
Q
What is adaptation to Norm and Tuning Curves?
A
- Opponent: Stare at features people are interested in = consequence is neuron populations tune to those features and less sensitive
- Adapted someone to facial features more extreme than norm, after adaptation they will lower their sensitivity
- Multichannel: Neurons that tuned to exact feature value would be affected and reduce sensitivity, as you move through, other neurons are not at all affected
- If you adapt to norm, opponent coding system = nothing should happen = norm is that stimulus = both neurons should activate equally, now both neurons lower sensitivity AND response, comparatively they are still the same
- MC: adapting to norm should have same effects as adapting to any other stimulus
- If you adapt to average face = little effect on subsequent perception
- Tuning function should be broadly tuned
18
Q
What is the distance between adaptor and target?
A
- Expect clear differences between both models as you change adaptor and target
- Adapted to two different anti-people
- Moved identities closer or further from the average face and wanted to know what effect on perceiving average face - moving should make them look more familiar, but if you move them further away from average = more effect on one population of neurones, and less on the other populations
- Further away = stronger effects should become
- MC: should expect the opposite effect = further = adaptation would have a lesser effect
- Aftereffect becomes larger as you move away from target = supports opponent system
