Abstract
Weiskrantz Saunders (1984) have shown that monkeys with bilateral lesions of the inferior temporal cortex (IT) are impaired in tasks that require the identification of an
object when it has been transformed in size or orientation. Subsequent research (eg. Logothetis et al, 1995) has shown that view-invariant representations can be set up in
IT neurons after extensive training.
We tested the hypothesis that macaques can learn view invariant recognition of novel objects in short periods of time. The responses of single neurons in the IT to four
digitised views of the object were tested before and after the object was shown continuously transforming between different views for periods of 30 s to 30 min. The
training period included manipulation of the object by the monkey as a result of which the monkey saw the object from different views. A novel control object was also included in the design, but in the “training” period only the four fixed views were shown, with no continuous transform between views.
Of the 103 neurons tested, 5 showed a significant change in their responses to the test object without any change in their responses to the control object (with interaction
terms in a two-way ANOVA significant in the range p<0.001 to p<0.05 for different neurons). These changes were apparent after test periods of 5-20 mins. We also
calculated an “invariance index” defined as the variance of the neuronal responses between views normalised over the mean firing rate. For the tested objects, the mean
invariance index over these 5 neurons changed from 3.07 to 1.40, indicating a more similar neuronal response to the different views after training, whereas there was no
significant change for the control object.
Thus some neurons in the IT have the ability to learn quickly (in less than 20 min) view invariant representations of novel objects.
