Artificial intelligence (AI) is concerned with building systems that simulate intelligent
behavior. It encompasses a wide range of approaches, including those based on logic,
search, and probabilistic reasoning. Machine learning is a subset of AI that learns to
make decisions by fitting mathematical models to observed data. This area has seen
explosive growth and is now (incorrectly) almost synonymous with the term AI.
A deep neural network is one type of machine learning model, and when this model is
fitted to data, this is referred to as deep learning. At the time of writing, deep networks
are the most powerful and practical machine learning models and are often encountered
in day-to-day life. It is commonplace to translate text from another language using a
natural language processing algorithm, to search the internet for images of a particular
object using a computer vision system, or to converse with a digital assistant via a speech
recognition interface. All of these applications are powered by deep learning.
As the title suggests, this book aims to help a reader new to this field understand
the principles behind deep learning. The book is neither terribly theoretical (there are
no proofs) nor extremely practical (there is almost no code). The goal is to explain the
underlying ideas; after consuming this volume, the reader will be able to apply deep
learning to novel situations where there is no existing recipe for success.
Machine learning methods can coarsely be divided into three areas: supervised, unsupervised, and reinforcement learning. At the time of writing, the cutting-edge methods
in all three areas rely on deep learning (figure 1.1). This introductory chapter describes
these three areas at a high level, and this taxonomy is also loosely reflected in the book’s
organization.