ABSTRACT
The scaffolded network is a network of neural networks whose objective is to test the hypothesis that the learning of complex concepts requires both a developmental progression and multiple perspectives of input data. The scaffolded network model incorporates three kinds of basic neural network architectures: a recurrent cascade net, a Kohonen net, and a recurrent Elman network and is motivated by physiological and psychological models. It is tested by teaching it simple mathematical concepts and functions and then comparing output and intermediate results with studies from developmental psychology. This design attempts to extend neural network capabilities to a more robust approximation of the cognitive phenomenon of cumulative learning.
