Summary

In this chapter, we started with the basic multilayer perceptron network. From there, we have talked about the basic structures, such as the input/output layers as well as various types of activation functions. We have also given detailed steps on how the network learns with the focus on backpropagation and a few other important components. With these fundamentals in mind, we introduced three types of popular network: CNN, Restricted Boltzmann machines, and recurrent neural networks (with its variation, LSTM). For each particular network type, we gave detailed explanations for the key building blocks in each architecture. At the end, we gave a hands-on example as an illustration of using TensorFlow for an end-to-end application. In the next chapter, we will talk about applications of neural networks in computer vision, including popular network architectures, best practices...