Why let networks grow?

This chapter examines more directly how development can be conceptualized within the context of neural networks that people learn. It argues that experience-dependent architectural plasticity has been largely under-emphasized in current accounts of neural network learning. It uses the example of auditory localization in barn owls as a case study in which architectural adaptation plays a fundamental role. The presence of architectural plasticity also has consequences for models of higher-level cognitive abilities. It shows how the cascade correlation learning architecture can be used to model a broad range of developmental phenomena in children's reasoning. In particular, it argues that architectural plasticity may underlie what Piaget described as stages of development.