I found this book by Andrew Trask back when I had first built my PC for grad school. I tried it as an epub, but the code and diagrams were unreadable.
Then I got the PDF version, and it was somewhat interesting. I had heard the hype, and going through the examples wasn't that bad. I'm less than a year total from a BSAE and BSME at UCF, and got my BS in Mechatronics Engineering from NC State, as well as joined GA Tech's OMSCS program. I had gotten a Math Minor at the Joint Program, and I was familiar with Matrix Operations due to taking Numerical Methods classes at UCF, and then applying those techniques in subjects such as FEM/FEA, CFD, and Heat Transfer.
I'm trying to say that I have quite a deep understanding of Ordinary and Partial Differential Equations, Linear Algebra, Difference Equations (The linearization of O/PDEs), and making valid simulations out of them, as well as things like Upper and Lower Triangular forms, Single Value Decomposition, etc. for making the computation easier (hey, I took this class in 2006!)
Then, there was a section that was showing different ways of expressing the same concept. One was a line of Python code, another was just shaped boxes, but the one that grabbed me and sent that special bolt of (eureka + over a decade of understanding all at once) was the line explaining the network in Linear Algebra terms. At that point, I wasn't confused any longer. I knew exactly what was going on, I knew exactly why we needed to use nonlinear activation layers. I knew it in a way only somebody who'd...well...grokked LinAlg would know.
That's not to say that I haven't had problems with the book. I've messed up, I've tried to wrap my head around concepts and felt like I wrapped around a brick instead.
But that was like trying to understand a new Genre. I already know the alphabet, the grammar, the vocabulary and lexicon.
I understand it's cluttered. I'm a very right brained thinker, and when I coded in school I used to flesh my ideas out with a mechanical pencil and printer paper. Both tools that I lack because having a Mechatronics degree (STEM) isn't what the news says it is.