Adam Pitko

Instytut Psychologii, Uniwersytet im. A. Mickiewicza
Szkota Wyższa Psychologii Społecznej


Abstract. This paper presents connectionist models of cognitive development, which
yield stage-like behavioural properties, but are based on static or generatiye neural
networks. These networks differ in learning rules that they use. Static networks use
only one kind of mechanism of change in information processing that consists in quanti-
tative changes of units connectivity. Generative networks can use more powerful kind
of mechanism of change that consists in quantitative as well qualitative changes in net
structure by recruitment of new units. It is discussed how connectionist approach can
provide insight into the role that an environmental input and learning constraints play
in the cognitive development. Simulation models of the acquisition of English past tense
and the modular model of the development of seriation are described that help illus-
trate the problem of quantitative and qualitative mechanisms of change. The article
concludes that different types of developmental changes can result from the interaction
between the specific requirements of task (e.g. conflicting types of input-output
mappings) and learning constraints.

Autor: Ewelina Soszyńska
Ostatnia aktualizacja: 11.03.2007, godz. 17:59 - Ewelina Soszyńska