on sample complexity and scaling issues of gradient descent trained multilayer perceptrons and a novel object-oriented simulation framework for scalable information processing modules