Visual categorization shapes feature selectivity in the primate temporal cortex

1
Visual categorization shapes feature selectivity
in the primate temporal cortex

• Natasha Sigala
• Nikos K. Logothetis
• In Nature, vol 415, 17 Jan 2002

2
Questions

• Generally How do we categorize objects? What
  does the Inferior Temporal (IT) cortex do?
• Which object features are represented in neurons
  in the inferior temporal cortex of the macaque?
• How are these representations affected by
  categorization training?
• Specifically Do Inferior Temporal neurons fire
  differentially in response to object features
  that are experimentally deemed to be significant
  for success on a categorization task?

3
Alternatives

• Neurons in the IT cortex will differentially
  respond to specific features after categorization
  training
• Neurons in the IT cortex will not differentially
  fire in response to specific features after
  categorization training

4
Logic

• Experience effects our perception of a given
  object
• Neurons in IT are known to be part of the ventral
  stream-- what processing system
• If experience effects the way we perceive
  objects, neuronal firing will show itthat is,
  experience of attention to specific features will
  change the firing patterns for objects that we
  see subsequently

5
Methods

• Two happy macaque monkeys
• Stimuli parameterized line drawings of faces
  and fish (stimuli had measurable features)
• Diagnostic and non-diagnostic features
• Training
• Testing

6
Methods Stimuli

• Faces
• EH, eye height
• ES, eye separation
• NL, nose length
• MH, mouth height
• EH, ESseparable diagnostic
• NL, MH not separable non-diagnostic

7
Methods Stimuli

• Fish
• DF, shape of dorsal fin
• T, tail
• VF, ventral fins
• M, mouth
• DF, T separable diagnostic
• VF, M not separable non-diagnostic

8
Stimuli
9
Methods

• Categorization task
• Monkeys trained to pull a different lever
  depending on if are shown stimulus from category
  1 or 2
• Trained until they reached 98 accuracy in about
  ½ second (416 ms, 530ms)
• Tested34 face, 34 fish, 10 repetitions

10
Methods

• Recorded from 150 single neurons from anterior
  ITin different experimental sessions
• Of these 150, 96 responded to visual stimuli
• 96 tested with faces stimuli
• 65 tested with fish stimuli

11
Results

• 44 of 96 (45.8) neurons responded statistically
  significantly to one or more values of the face
  stimuli features
• 32 of these 44 (72.7) responded selectively to
  either one or both diagnostic features of the
  face stimuli (but did not respond for the
  non-diagnostic features)

12
Results

• 28 of 65 (43.1) neurons responded statistically
  significantly to one or more values of the fish
  stimuli features
• 21 of these 28 (75) responded selectively to
  either one or both diagnostic features of the
  fish stimuli (and not for the non-diagnostic
  features.)

13
Results Definitions

• There were three possible values for each
  stimulus feature (eg large, medium or small eye
  width)
• Best feature, defined as the stimulus value that
  provoked the largest number of spikes per second
• Worst feature, defined as the stimulus value that
  provoked the least number of spikes per second

14
(No Transcript)
15
(No Transcript)
16
Results Figure 4

• For both faces and fish, more neurons fired in
  response to the diagnostic features
• Red neurons that fired statistically
  significantly selectively for diagnostic features
  only
• Blue neurons that fired statistically
  significantly selectively for diagnostic and
  non-diagnostic features
• Black no significant selectivity

17
Interpretation

• This study supports the notion that perception of
  visual category information is processed in the
  Inferior Temporal cortex
• Inferior Temporal neurons do fire differentially
  in response to object features that are
  experimentally deemed to be significant for
  success on a categorization task.

18
Limitations

• Before and After snapshot missing It would have
  been an even more exciting study if the authors
  had shown the patterns of neuronal firing before
  the training and compared that to post-training
  firing rates
• Or alternatively, they could have trained other
  monkeys on the non-diagnostic features in order
  to provide more evidence that it was actually the
  training that caused the selective firing
• Having shown such an interesting result, it would
  have been even more theoretically important if
  they had correlated the monkeys category
  discriminations (behavior) with the patterns of
  neuronal firing.
• According to visual system expert (Pascal) the
  stimuli were a bit small for the monkeys to see