| |
Abstract:
Abstract: Objects in the natural world are commonly grouped
into categories. One difficulty in categorization is that visually
similar objects can belong to different categories while dissimilar
objects can belong to the same category. How do neurons process and
encode categorical information? We trained a monkey to perform a
delayed match-to-category task using 3D morphed objects spanning
two classes, 'cats' and 'dogs'. As the sample images approached the
class boundary, the monkey's performance fell sharply, suggesting
that it had indeed learned the categories. We recorded from 136
neurons in the ventrolateral PF cortex, of which 41% (n=56) showed
activity that reflected the category membership of the sample.
Their activity was similar for members of the same category but
different for stimuli that belonged to different categories.
Analysis of all recorded neurons revealed that a statistically
significant categorical signal was carried by the population.
During the choice phase of the task, nearly half (67/136) of
neurons responded differently to categorical matches vs.
categorical nonmatches, some showing match enhancement and others
showing match suppression. During error trials, there was a
substantial reduction in the category tuning of single neurons and
in the population. These findings suggest that prefrontal neurons
convey information about objects' class membership.
|
