Getting started with deep learning

Chapter 3. Major Architectures of Deep Networks

The mother art is architecture. Without an architecture of our own we have no soul of our own civilization.

Frank Lloyd Wright

Now that we’ve seen some of the components of deep networks, let’s take a look at the four major architectures of deep networks and how we use the smaller networks to build them. Earlier in the lesson, we introduced four major network architectures:

• Unsupervised Pretrained Networks (UPNs)
• Convolutional Neural Networks (CNNs)
• Recurrent Neural Networks
• Recursive Neural Networks

In this chapter, we take a look in more detail at each of these architectures. In Chapter 1, we gave you a deeper understanding of the algorithms and math that underlie neural networks in general. In this chapter, we focus more on the higher-level architecture of different deep networks so as to build an understanding appropriate for applying these networks in practice.

Some networks we’ll cover more lightly than others, but we’ll mostly focus on the two major architectures that you will see in the wild: CNNs for image modeling and Long Short-Term Memory (LSTM) Networks (Recurrent Networks) for sequence modeling.

Unsupervised Pretrained Networks

In this group, we cover three specific architectures:

• Autoencoders
• Deep Belief Networks (DBNs)
• Generative Adversarial Networks (GANs)

A Note About the Role of Autoencoders

As we previously covered in Chapter 2, autoencoders fundamental structures in deep networks because they’re often ...