Interactions between development and learning during the acquisition of binocular disparity sensitivities

Investigates the use of developmental progressions in the acquisition of binocular disparity sensitivity, In an earlier paper (2001) we presented results of simulations comparing a non-developmental neural network model and two developmental neural network models trained to detect binocular disparities and concluded that the developmental models consistently performed better. Here we report the results of simulations on additional neural network models, compare them to the three original models and perform additional analysis. We conclude that the benefits of development are not solely due to limiting the size of the input in early stages, and that in order for development to be helpful to the task of binocular disparity detection, the developmental progressions should be designed to take advantage of known simplifications.