FIG. 1.

A: visual recognition tasks. Three objects (star, triangle, cross) were shown at three possible positions (−2, 0, and +2° relative to the fovea) either in isolation or in combinations of pairs or triplets. Using the inferior temporal cortex (IT) population response data to each visual scene, linear discriminant classifiers were used to measure how well the population had solved 2 different visual recognition tasks. One task required the linear discriminants to classify object identity irrespective of its position (position-invariant task). In the particular example illustrated, the classifier was asked to classify the presence of a star (report yes to all visual displays that contain a star regardless of the star's position). In the other task, the classifier had to report object identity at a particular position (position-specific task). In the example illustrated, the classifier had to report yes only to the visual scenes in which the star was present in the top position while disregarding other displays (even those in which the star was present in another position). B: classification performance for a real IT population and a simulated V1 population on the position-invariant and -specific tasks. All performance was averaged performance using “leave-one-out” cross validation procedure (see details in methods).