Deep Learning I: Gradient Descent