Usually I'm a harsh critic of text like this because casual language and what look like casual words but are actually strictly defined domain specific technical definitions are utterly indistinguishable.<p>However! This book back-references its appendix with grey underlines thus signaling that something has a technical definition and is jargon.<p>For instance, "loss" is really common English. In ML it's a loss function. You can clearly see it's a special word in this world in the text. Other examples include weights, capacity and channel.<p>I don't have to sit there confused trying to guess which English words are being used in special ways.<p>This discipline is fantastic. In some texts, such terms might be italicized but that behavior has seem to fallen out of practice.<p>As someone who isn't a professional mathematician, hints like these help greatly.
I had a course named `Deep Learning` by François Fleure at EPFL, and he is a really an amazing professor, able to share his passion alongside explaining very advanced topics. It does not surprise me to see that his books are of very high quality !<p>The course was focusing on all of the mathematical aspects of Deep Learning, starting from the simple understanding of the gradient descent algorithm to (trying to) understand how transformers work. There was also quite a lot of computer science involved and lots of practical assignments. One of the 2 projects of this course was to design from scratch in Python or C++ a DNN framework (roughly an API like pytorch or tensorflow) which required to really think properly about which architecture to use for your code. The minimum requirements only asked to implements a few activation, normal layers and convolutional layers but you go beyond that and implements all kind of layers. Lots of fun. This course remains as one of my favorite courses.
By "little" I thought it would be brief, which I think it sort of is. But what it really means by little is that it is optimized for little screens such as a smartphone.<p>Go ahead and open it in your phone. You'll be delighted to read it.<p>Question for HN: how can I convert my existing PDFs and eBooks that I can easily read from my phone?<p>For e.g., I have a lot of Math textbooks in PDF format and I would like to convert them into a format similar to this deep learning book. How can I go about doing that?
His website also has a version for printing out on 36 sheets of paper and folding into a real book:<p><a href="https://fleuret.org/public/lbdl-a5-booklet.pdf" rel="nofollow">https://fleuret.org/public/lbdl-a5-booklet.pdf</a>
What a neat little book! I suppose this was typeset in TeX. How did they optimize for smaller width though?<p>Edit: I found the style templates on author's webpage [1].<p>[1]: <a href="https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?p=littlebook.git;a=summary" rel="nofollow">https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?p=littlebook.g...</a>
Great writing! I love how this book gets to the point right from the start, answering one of my questions about deep learning (how did this start?) in the very first sentence:<p>"The current period of progress in artificial intelligence was triggered when Krizhevsky et al.[2012] showed that an artificial neural network (..) could beat complex state-of-the-art image recognition methods by a huge margin (..)"
This is such a amazing little book. I love it, AI is hard for me to understand but this book makes it very easy to grasp some of the concepts.<p>Thanks for sharing