Learning

1
Learning Memory

• 6. Perceptual learning

2
Perceptual learning
Expertise-based inversion effects

• Essential
• Goldstone (1998)
• Wills, Suret, McLaren (2004)
• Suppl
• McLaren Mackintosh (2000)

McLaren Mackintosh model
Neural representation of faces 2021-3
Caricatures
When exposure goes bad!
Prosopagnosia 2021-3
Feature creation?
Before
After
During
3
Neural representation of faces
Tanaka (1993)
Inferotemporal cortex (IT)
Face-specific cells? Dave Perret
Instance cells Note relation to previous lecture!

• Faces, like many other objects, have
  instance-based representations in IT.

4
Prosopagnosia
Farah (1995)

• Brain damage can affect expert categories more
  than inexpert categories.

5
Inversion effects
Phase 1 Inspection series
Phase 2 Forced-choice
.........

• Half appear upright at inspection and test, half
  inverted at inspection and test

6
Inversion effects

• Diamond Carey (1986)
• Picture types
• Photographed faces.
• Photographed Irish Setters and Cocker Spaniels.
• Subject types
• American Kennel Club listed breeders of Setters
  and Spaniels with an average 31 years of
  experience.
• Undergraduates.
• Procedure basically as Yin (1969)

7
Dog inversion effect

• Inversion can affect expert categories more
  than non-expert categories.

8
Perceptual learning

• Faces, and other difficult, expert, categories
  undergo perceptual learning as a result of our
  extensive exposure to them.
• Perceptual learning Where exposure to stimuli
  makes them easier to discriminate from each
  other.
• How does it happen?

9
Perceptual learning Mechanisms

• McLaren and Mackintosh (2000, 2002)
• Associability Rate at which a feature forms
  associations to other representations (e.g. the
  persons name)
• All stimuli are composed of multiple features.
• When stimuli are repeatedly presented, the
  associability of their features begins to drop.

10

• Faces have a basic, prototypical form.
  Individuals are minor random variations.

11

• Faces are composed of many common features, and a
  few unique features

Common
Unique
12

• We have a lot of experience of faces
• That reduces the associability of all features

Common
Unique
13

• The associability of the common features reduces
  more because they have occurred more often
• This is good news for discriminability, because
  its the unqiue features that are helpful here

14
Dick
Dick
Tom
Tom
Hari
Hari
Beth
Beth
Sue
Sue
Liz
Liz
15
Face inversion effect

• Familiarity-based inversion effect
• Familiarity improves discrimination performance
  through perceptual learning.
• Inversion disrupts and destroys familiar features
  - advantage brought by familiarity is lost.

16
Which is Ronald Reagan?
17
Automatic caricature
Photo
Veridical
Caricature
Anti-caricature

• Brennans (1985) program automatically generates
  caricatures by exaggerating differences between a
  veridical drawing and a norm face which is the
  average of a large number of faces.

18
Rhodes, Brennan Carey (1986)

• Speeded naming of line drawings of faces familiar
  to subjects
• Naming R.T. (secs)
• Anti-caricature 12.5
• Veridical 6.0
• Caricature 3.2
• The caricature process increases the salience of
  unique features, enhancing the effect of
  perceptual learning.

19
Stronger evidence for McL M

• If the task is to discriminate two categories
  (e.g. cats and dogs) and,
• If you could have a category where
• The diagnostic (unique) features,
• were just as frequent as the non-diagnostic
  (common) features
• Then exposure would make you worse!
• Detrimental effect of lowered associability
• No beneficial effect of differential
  associability

20
Wills, Suret McLaren (2004)
21
Feature creation?

• McL M model is about changing the associability
  of pre-existing features.
• Might exposure also cause the creation of new
  features?

22
Schyns Rodet (1997)
A 19
Category A
Category B
Category C
Category ?
A 88
Category B
Category A
Category C
23
What next?
Atkin Shif 1117,1121-3
Glanzer Cunitz 1121-3
Essential Goldstone (1998) Wills, Suret,
McLaren (2004) Suppl McLaren Mackintosh (2000)
Word length effect 1121-3
PhonoSem sim 1121-3
STM HM vs KF 1121-3
Intro WMM 1121-4
VSSP dissoc 1121-4
WM - Neurosci 2021-7