I need your help!
Folks, I really need to hear your thoughts and opinions on a small issue.
I've been writing a book called "Mastering Deep Learning: A No Maths Introduction to Contemporary Artificial Intelligence" and I'm now discussing with a publisher about whether there is a demand for such a book. If I explain what the book is about, I'd really appreciate your thoughts on whether an audience for it.
Current AI books, I think, fall into three types. First, heavily mathematical or programming books. Second, management guides to applications of the technology. And, finally, "we are all doomed" books.
I believe there is a demand for a new type of book.
It is not new to show what AI (and DL and ML) can do - that appears frequently in the newspapers. What is groundbreaking is to show how it does it - in a way that means normal people (ie those who flee from any maths which contains more letters than numbers) can understand.
The aims of the book are:
- To enable people to develop deep learning models using graphical tools
- To give them a platform from which to investigate the subject more deeply
- To be able to critically appraise media reports about DL and AI
The book structure is:
Introduction - high profile applications of deep learning: autonomous cars, defeating the best GO player in the world and so on.
Chapter One – The Black Box
This chapter uses an extended metaphor of neural networks as black boxes to help introduce important concepts underlying deep learning, including space, decision boundaries, Voronoi diagrams, cost and loss, dimensionality reduction, and learning as movement in space.
Chapter Two – Fully Connected Networks
This chapter looks inside the black box to see how neural networks function. This chapter focuses of the “simplest” approach to deep learning which use simple processing nodes, organised into layers with every node in each layer connected to every node in both the preceding and succeeding layers.
Chapter Three – Seeing: Computer Vision
This chapter starts by defining the range of the domain of computer vision as being the answers to the question - “what questions could we ask about this photo?”. These range from “is there a cat in the picture?”, through “point to the cat” to “is that fuzzy thing in the corner a cat or a dog?”.
Chapter Four – Doing: Reinforcement Learning
After a discussion of the characteristic of games, this chapter introduces the concept of Markov decision processes. This is used to describe the notion of reinforcement learning. The Bellman Equations are introduced (using a metaphor) and they are used to describe the many different approaches to RL.
Chapter Five – Understanding: Natural Language Processing
This chapter looks at deep learning approaches to language and compares these to traditional linguistic approaches.
Chapter Six – Imagining: Generative models
This chapter takes the book in a new and creative direction – can we use deep learning to produce interesting art? This was mentioned regarding music in last chapter and now we will look at image generation.
Chapter Seven – Practical Deep Learning
This chapter moves from understanding deep learning to actually doing some DL projects. It covers three main topics: how to structure a DL project, tools to help develop these projects, and where to find more information to continue learning the subject.
Chapter Eight – Conclusion
In this final chapter we briefly review the key concepts learnt in this book and critically look at recent AI scare stories.
So what do you think? Would people buy this book? What kinds of people? Would you be interested in reading this book?
Please let me know - even if you think the answer to these questions is "no".
Hi Steven, I would buy this book. As a non-technical person in business, I would find it useful to understand the fundamentals of AI, which seem like it will only become a more prevalent technology in our lives. I think it would also be useful to understand what are the capabilities and limitations of AI now and in the near future, especially before introducing executing deep learning projects, and to communicate with technical people.
I really like the title and concept. Whilst important to recognise that ML is essentially very complex applied mathematics, there is a great deal that scientifically/technically minded person can learn on the topic by intuition, that isn't enabled by attempting to parse multi-line equations/derivations in text-book style. This sounds like the correct approach to me.
I've been reading an ML book that is helpfully spending time on talking about the results of ML calculations & whether you get a good result, or just an obvious result that you might have found with a much simpler algorithm. That's not a topic I've seen so much myself in other books, but it strikes me as important for managing investment into ML development for applications - you don't always get value from taking the more complex approach.