I'm currently working on my third book, a book about neural networks and deep learning. Here's the first chapter:

http://neuralnetworksanddeeplearning.com/chap1.html

It'll be free online once complete, but I raised money with an Indiegogo campaign (about 10k). I'm self-publishing, in part because I wanted full control over the book. This gives me the ability to experiment with in-text videos, and with other tricks. For instance, when a reader clicks on an equation reference in the text, the relevant equation appears in the margin, as a reminder. Clicking on the marginal equation will take you to the context in which the equation originally appeared. This cuts down on tedious back-and-forth.

My two earlier books were both published in the traditional way:

+ A book for general audiences about networked science, "Reinventing Discovery: the New Era of Networked Science": http://www.amazon.com/dp/0691160198 Published by Princeton University Press.

+ A textbook about quantum computing, jointly with Ike Chuang: http://www.amazon.com/Quantum-Computation-Information-Annive... Published by Cambridge University Press. No e-book at all when first published (2000)!

Both were written with LaTeX, and included many illustrations. "Reinventing Discovery" was actually rekeyed entirely by the publisher, but the textbook was produced from our LaTeX copy.

I would recommend the book mentioned at the bottom as well -- http://www.amazon.com/Reinventing-Discovery-The-Networked-Sc...