Modelling cognitive developmental transitions in neural networks: bifurcations in an adaptive resonance theory model

This chapter adopts a different solution to the question of stage transitions in development. It emphasises the dynamic nature of the neural information processing that underlies cognitive development. New evidence examining the details of children's responses suggests that many of them show 'catastrophe flags' characteristic of such transitions at critical developmental junctions. The discussion shows that these kinds of transitions can take place in recurrent neural networks in which activation is allowed to spread in all directions. It demonstrates this using an adaptive resonance theory (ART) network to model discontinuities in concept learning.