| |
Abstract:
Although the ability to group stimuli into meaningful
categories is a fundamental cognitive process, almost nothing is
known about its neural basis. To explore this, we trained monkeys
to categorize computer-generated stimuli as "cats" and "dogs". A
novel graphics morphing system was used to systematically vary
stimulus shape and precisely define an arbitrary category boundary.
We found that the activity of many neurons in the lateral
prefrontal cortex (PFC) reflected the category of visual stimuli
independent of stimulus similarity, and that these categorical
representations changed during learning when a monkey was retrained
with the same stimuli assigned to new categories. Psychophysical
analysis of the monkeys behavior and the analysis of eye movements
suggest that the monkeys solve the task by attending to multiple
features of the stimuli. Removing any single feature of the
stimuli, such as the heads or tails, caused task performance to
drop slightly but remained significantly above chance in both
cases. This demonstrates that the monkeys can learn to use a
conjunction of stimulus attributes to make categorical judgements.
Furthermore, this suggests that categorical tuning of PFC neurons
is not due to simple discrimination of single stimulus features but
rather arises from the perceptual interaction of multiple object
features, a hallmark of categorization.
|
