This is Chapter 3 of the book titled "Deep Learning": a nine-part easy-to-grasp textbook written with the goal of demystifying the science behind contemporary AI. The book attempts to explain the fundamentals of Deep Learning to readers from a wide range of backgrounds in an intuitive manner, while navigating its strong mathematical foundations in a highly pictorial style. This chapter is also the first in the series of chapters under Part II of this book, dealing with neural network training. In Chapters 1 and 2, we saw that any function can be modeled by an appropriately architected neural network. However, for the network to model a function accurately, its parameters must be properly assigned. In this chapter we discuss the problem of how the parameters of a network can be learned from training data, and how the learning can be cast as an optimization problem that can be solved through the method of gradient descent.