Representation of objective similarity among three-dimensional shapes in the monkey.

Abstract

Humans have been recently shown to represent parameterized three-dimensional objects in a manner that preserves relative similarities (as measured by parameter-space distances) among the objects (Cutzu and Edelman 1996). We show that the representation of objects in the monkey visual system is similarly faithful to the parametric variation built into the stimulus set. A monkey (Macaca fuscata) performed a delayed matching-to-sample task with 28 images (4 views x 7 objects). Stimuli in each of the two experiments were seven computer-rendered parameterized animal-like shapes, arranged in a low-dimensional configuration (namely, a two-dimensional TRIANGLE) in a common 56-dimensional parameter space. The monkey's task was to match objects (not views). Each experiment lasted for 3-4 weeks after the introduction of the stimulus set to the subject. Error rates were entered into a 7 x 7 object confusion matrix and submitted to nonmetric multidimensional scaling (MDS). In both experiment 1 (mean correct rate 69.7%) and experiment 2 (mean correct rate 59.9%), the MDS solutions resembled closely the low-dimensional parameter-space patterns built into the stimuli, in the sense that the point corresponding to the central one in the original pattern was inside the other six points and that the order of the six points in angular positions around the center point was preserved. A simulation study showed that the resemblance could not be due to chance. These results demonstrate the possibility of veridical representation of parametric similarity among complex objects in the monkey.