Learning a Specialization for Face Recognition : The E ect

The double dissociation between prosopagnosia, a face recognition deecit occurring after brain damage, and visual object agnosia, diiculty recognizing other kinds of complex objects, indicates that face and non-face object recognition may be served by partially independent mechanisms in the brain. Such a dissociation could be the result of a competitive learning mechanism that, during development, devotes neural resources to the tasks they are best at performing. Studies of normal adult performance on face and object recognition tasks seem to indicate that face recognition is primarily conngural, involving the low spatial frequency information present in a stimulus over relatively large distances, whereas object recognition is primarily featural, involving analysis of the object's parts using local, high spatial frequency information. In a feed-forward computational model of visual processing, two modules compete to classify input stimuli; when one module receives low spatial frequency information and the other receives high spatial frequency information, the low-frequency module shows a strong specialization for face recognition in a combined face identiication/object classiication task. The series of experiments shows that the ne discrimination necessary for distinguishing members of a visually homogeneous class such as faces relies heavily on the low spatial frequencies present in a stimulus. 1 Background Studies of the preserved and impaired abilities in brain damaged patients provide important clues on how the brain is organized. Cases of prosopagnosia, a face recognition deecit often sparing recognition of non-face objects, provide some evidence that face recognition is served by a \special" mechanism. What might the basis for this dissociation be? The fact that visual object agnosia, an object recognition deecit, can occur without appreciable impairment of face recognition 1] precludes the possibility that face recognition is simply harder than object recognition but is served by the same mechanism. Another possibility, that prosopagnosics have impaired within-category discrimination but intact superordinate classiication, can be rejected on the basis of studies of extreme prosopagnosic patients such as W.J., who learned a diicult sheep discrimination task remarkably well but could not learn new faces 11]. In previous work, we proposed that a neural mechanism allocating resources according to their ability to perform a given task could explain the apparent specialization for face recognition evi-denced by prosopagnosia 4]. We showed that a model based on the mixture of experts architecture 10], in which a gating network implements competitive learning between two simple homogeneous modules, could develop a specialization such that damage …