This chapter is from the book
This chapter is from the book
This chapter is from the book
Summary
We covered a lot of ground in this chapter. Starting from an appreciation of how a neural network with fixed parameters processes information, we developed an understanding of the cooperating methods—cost functions, stochastic gradient descent, and backpropagation—that enable network parameters to be learned so that we can approximate any y that has a continuous relationship to some input x. Along the way, we introduced several network hyperparameters, including learning rate, mini-batch size, and number of epochs of training—as well as our rules of thumb for configuring each of these. The chapter concluded by applying your newfound knowledge to develop an intermediate-depth neural network that greatly outperformed our previous, shallow network on the same handwritten-digit-classification task. Up next, we have techniques for improving the stability of artificial neural networks as they deepen, enabling you to architect and train a bona fide deep learning model for the first time.