Neonatal Learning of Faces

Newborn face perception is controversial, but the current evidence suggests that (a) newborns follow face-like schematic patterns further than similar patterns, (b) infants can learn individual faces soon after birth, and (c) full face processing abilities develop through months or years of experience with faces. Previous models have not adequately accounted for all three types of results. In prior work, we showed how a biologically based self-organizing system and spontaneous activity patterns can explain newborn face preferences. In this paper we show that this general-purpose learning system can explain both neonatal and later learning. Using computational simulations, we demonstrate that newborn learning need not be based on the external outline, as has been supposed, and that postnatal decreases in response to schematic faces need not represent a decrease in response to real faces. These simulations provide concrete predictions to guide future experiments with infants, while suggesting new techniques for designing complex adaptive systems in general.