Age of Acquisition in Facial Identification: A Connectionist Approach

Age of Acquisition (AoA) is the phenomenon that acquiring a certain piece of information earlier than another results in a faster response time in adulthood. AoA has been shown to have a significant role in a variety of human studies. Recently, it has been demonstrated that connectionist networks that abstractly model reading and arbitrary mappings can also show AoA effects, and we extend this to facial identification. We present a connectionist model of facial identification that demonstrates strong AoA effects by allowing faces to be acquired in their natural order and by staging face presentation. This extends previous work by showing that a network that simply classifies its inputs also shows AoA effects. We manipulate the staged model in two ways, by either assuming outputs for the late set are trained to be off early in learning (or not) and by assuming the representation developed for the early set is used for the late set (or not). In three of these cases, we find strong AoA effects, and in the fourth, we find AoA effects with a recency control.