chapter 3: improving the way neural networks learn