Part IV. Deep Unsupervised Learning using TensorFlow and Keras

Until now, we have worked with only shallow neural networks; in other words, networks with only a few hidden layers.

Shallow neural networks are certainly useful in building machine learning systems, but the most powerful advances in machine learning over the past decade have come from neural networks with many hidden layers, known as deep neural networks.

This subfield of machine learning is known as deep learning. Deep learning on large, labeled datasets has led to major commercial successes in areas such as computer vision, object recognition, speech recognition, and machine translation.

We will focus on deep learning on large, unlabeled datsets, which is also commonly referred to as deep unsupervised learning. This field is still very new, full of potential but with fewer commerical successes to date compared to the supervised variant.

Over the next few chapters, we will build deep unsupervised learning systems, starting with the simplest building blocks.

In Chapter 10, we will explore restricted boltzmann machines and use these to build a recommender system for movies.

In Chapter 11, we will stack restricted boltzmann machines on top of each other, creating deep neural nets known as deep belief networks.

In Chapter 12, we will generate synthetic data using generative adversarial networks, one of the hottest areas of deep unsupervised learning today.

In Chapter 13, we will return to clustering but work with time series data this time.

This is a lot of advanced material, but you will see that a lot of deep unsupervised learning draws on the fundamental unsupervised learning principles we introduced earlier in the book.