Beyond backprop: emerging trends in connectionist models of development: an introduction

Abstract

Since the publication in 1986 of Rumelhart and McClelland’s Parallel distributed processing: Explorations in the microstructure of cognition , neural network or connectionist models have become an increasingly common method for studying learning and development. A wide range of developmental domains have been investigated with connectionist models, including language acquisition, perceptual development, object permanence, developmental psychopathology and motor skill acquisition. Many of these models rely on the backpropagation-of-error learning algorithm, a form of supervised learning in which a ‘teacher’ shapes the output of the network by providing it with desired responses.