Gradient descent, how neural networks learn | Deep learning, chapter 2