Modeling the development of causality and occlusion perception in infants

Developmental researchers investigate many pieces of infants’ physical knowledge, e.g. the perception of causality, occlusion or object permanence, but a theoretical framework that would unify all these pieces, account for the most basic phenomena and make testable predictions has not been provided yet. Here we make an attempt to unify and explain the emergence of causality and occlusion perception and its development in infancy using a simple artificial neural network that derives its representations from simplified motion detector and disparity cells as found in the primary visual cortex. The network accounts simultaneously for two experiments on causality and occlusion perception and develops a representation of object permanence during training. It also makes detailed testable predictions for the course of development and provides an account of how change occurs. We conclude that many aspects of physical knowledge can probably be learned from the statistical regularities of our environment while only few assumptions are needed.