Internal representation models in feedforward artificial neural networks