Programmed cell death during early development of the nervous system, modelled by pruning in a neural network

Autor: Vos, JE, vanHeijst, JJ, Greuters, S, Silva, FL, Principe, JC, Almeida, LB
Přispěvatelé: University of Groningen
Jazyk: angličtina
Rok vydání: 1997
Zdroj: SPATIOTEMPORAL MODELS IN BIOLOGICAL AND ARTIFICIAL SYSTEMS, 192-199
STARTPAGE=192;ENDPAGE=199;TITLE=SPATIOTEMPORAL MODELS IN BIOLOGICAL AND ARTIFICIAL SYSTEMS
Popis: An artificial neural network model is presented in which the development is simulated of a baby's ability to control movement of his forearm around the elbow, until he is capable of goal-directed reaching. The neural network implementation provides the facility to change the number of nodes (or artificial neurons), and thus allows simulating 'natural' or 'programmed' cell death in the nervous sytem and its effects on learning in, and performance of, the motor control system. It was found that a gradual decrease in the number of artificial interneurons could be allowed once the network's task had been learned. Every time some 'relearning' was necessary, which implied that the network made small adjustments in its connectivity. When it was tried to train the network with only few interneurons right from the beginning, the task could never be learned. In the discussion an attempt is made to understand the biological relevance of the initial overproduction and the subsequent death of a large fraction of the number of neurons.
Databáze: OpenAIRE