Cascade-correlation learns the balance scale task
During my Masters I replicated the study "Modeling Cognitive Development on Balance Scale Phenomena" of Thomas R. Shultz et al. I did this for the course "Simulation of Growth and Development" from Han van der Maas. The study tried to model the development of children which learned to perform a balance scale task. They had to predict to which side the balance would tip given weight on certain positions on the scale. The model involved a specific artificial neural network method called cascade correlation.
I wrote a report in dutch: gzipped postscript (137 kb), pdf (230 kb).
The difficult part in replicating this study, was implementing the cascade correlation method, because not all details of its implementation were given. If you dare to look at it, here is my code: cascadebalancecode.tgz (11kb) (owned by the University of Amsterdam).
The results of my project were used in a much bigger research project by Philip Quinlan from the University of York and Han van der Maas, which were both also working on the same topic. This resulted in the paper "Re-thinking stages of cognitive development: An appraisal of connectionist models of the balance scale task ".
