Deep Learning: A Visual Approach

Book Preface

Imagine that you’re rubbing a golden lamp. You say, “Genie, for my three wishes, give me someone to love, great wealth, and a long and healthy life.”

Now imagine that you’re entering your home. You say, “House, bring the car around, ask Sarah if she’s free for lunch, schedule a haircut, and make me a latte. Oh, and play some Thelonious Monk, please.”
In both of these situations, you’re asking a disembodied being of great power to hear you, understand you, and fulfill your desires. The first scenario is a fantasy going back millennia. The second scenario is commonplace reality today, thanks to artificial intelligence, or AI.
How did we invent these magic genies? The revolution in AI that’s changing the world today is the result of three developments.

First, computers are getting bigger and faster every year, and special-purpose chips originally intended for generating images are now powering AI techniques as well.
Second, people keep developing new algorithms. Surprisingly, some of the algorithms powering AI today have been around for decades. They were more powerful than anyone knew and were just waiting for enough data and compute power to let them shine. Newer and even more powerful algorithms are now pushing the field forward at an increasingly rapid pace. Some of the most powerful algorithms used today belong to a category of AI called deep learning, and those are the techniques we’ll emphasize in this book.

Third, and perhaps most critically, there are now massive databases for these algorithms to learn from. Social networks, streaming services, governments, credit card agencies, and even supermarkets measure and retain every crumb of information they can gather. Public data on the web is also vast. As of late 2020, estimates suggest there are more than 4 billion hours of video freely available online, over 17 billion images, and an untold amount of text covering topics from sports history to weather patterns to municipal records.

In practice, these databases are often seen as the most valuable component of any new machine learning system. After all, anyone with money can buy computers, and the algorithms they run are almost all publicly available in research journals, books, and open source repositories. It’s the data that organizations jealously hoard, to use for themselves, or to sell to the highest bidder. It’s no stretch to say that databases are the new oil, the new gold.

Like any radical technology, AI has its rosy cheerleaders who foresee great advantages for the human race and dour predictors of doom who see nothing but ruin and the destruction of the societies and cultures that make up our world. Who is right? How can we judge the risks and rewards of this new technology? When should we embrace AI and when should we forbid it?

The best way to thoughtfully deal with a new technology is to understand it. When we know how it works, and the nature of its powers and limitations, we can determine where and how it should be used to create a future we want to inhabit. This book exists to help you understand what deep learning is and how it works. When you know the strengths and weaknesses of deep learning, you’ll be in a better position to use it for your own work and understand its actual and potential impacts on our cultures and societies. It will also help you see when people and organizations that hold power are using these tools, and then determine for yourself whether they’re using them for your advantage, or their own.

Who This Book Is For

I wrote this book for anyone who’s interested in how deep learning works. You don’t need math or programming experience. You don’t need to be a computer whiz. You don’t have to be a technologist at all!

This book is for anyone with curiosity and a desire to look behind the headlines. You may be surprised that most of the algorithms of deep learning aren’t very complicated or hard to understand. They’re usually simple and elegant and gain their power by being repeated millions of times over huge databases.
In addition to satisfying pure intellectual curiosity (which I think is a fine reason to learn anything), I wrote this book for people who come face to face with deep learning, either in their own work or when interacting with others who use it. After all, one of the best reasons to understand AI is so we can use it ourselves! We can build AI systems now that help us do our work better, enjoy our hobbies more deeply, and understand the world around us more fully.
If you want to know how this stuff works, you’re going to feel right at home.

This Book Has No Complex Math and No Code

Everybody has their own way of learning. I wrote this book because I felt there were people who would like to understand deep learning but didn’t want to get there by studying equations or programs.
So, rather than describe algorithms with equations or program listings, I use words and pictures. That doesn’t mean the descriptions are sloppy or vague. Rather, I’ve worked hard to be as clear as possible, and, when appropriate, precise. When you’re done with this book, you’ll have a solid grasp of the general principles. If you later decide that you’d like to reframe your understanding in mathematical language, or write programs with a specific computer language or library, you’ll find it much easier than starting from scratch.