Visual Expertise is a General Skill

The fusiform face area (FFA) in the ventral temporal lobe has been shown through fMRI studies to selectively respond with high activation to face stimuli, and has been identified as a face specific processing area. Studies of brain-lesioned subjects with face recognition or object recognition deficits also have often been cited as evidence for face specific processing. Recent studies, however, have shown evidence that the FFA also responds with high activation to a wide variety of non-face objects if the level of discrimination and the level of expertise are controlled. Based on these recent results, we hypothesized that the features of faces that the FFA respond to can be useful for discriminating other classes of visually homogeneous stimuli with some tuning through experience. To test our hypothesis, we trained two groups of feed-forward neural networks on visual classification tasks. The first group was pretrained on basic level classification of four stimulus classes, including faces. The second group was pretrained on subordinate level classification on one of the stimulus classes and basic level classification on the other three. In two experiments that used different criteria to stop pretraining, we show that networks that fully acquire the skill of subordinate level classification consistently show an advantage in learning the new task.