Abstract
Face perception is likely influenced by experience, and people may frequently encounter groups of relatively similar faces, e.g. from the same family, or 'race.' Can observers learn to establish visual representations for groups of faces, making groups that initially appear similar more distinct? We tested this idea by training observers on two groups of faces and measuring perceived similarity. Groups of 5 faces were defined using an established 'face space' that represents differences in geometric features as distance. Faces within a group were close to each other, compared to randomly selected faces, and the two groups were also relatively close. Human observers then viewed two faces at a time and rated, from 1-5, how similar the pairs were, for all possible pairings of the 10 faces. We next trained observers to identify members of the groups using a visual search task. Target and distractor faces were chosen from the same group, and the two groups were trained in alternate sessions. Observers completed two visual search sessions a day for three consecutive days. Following training, we again measured perceptual similarity. Overall, face pairs from the same group were rated as more similar than pairs from different groups. Faces appeared less similar following training; the mean similarity rating for pairs within the same group increased significantly, (m=.91, p<0.01). Critically, the mean rating for pairs from different groups also increased (m=.56, p<0.01). A 2D multidimensional scaling solution additionally showed that the distance between the centroids of the two groups increased (by 0.36). These results suggest that experience can alter how our visual system represents groups of faces, decreasing similarity both within and between groups. Bringing the appropriate group-level representation to bear on frequently-encountered groups may improve performance on tasks such as identification or recognizing expressions.