Why?

I kept finding myself drawn back toward AI/ML research around this period. There are probably several reasons why.

• mild ADD means I constantly have to have a new thing to latch on to and try out. Yes, I’ve tried to get on medication (there’s a shortage nationwide and I don’t feel like hitting up Tor to get a fix of whatever they’re whipping up) but I’d rather just hyperfixate on a wide array of topics on rotation every couple months. There are worse habits.
• it’s what the cool kids are working on
• C.R.E.A.M.

https://www.youtube.com/watch?v=4yNQ7_7I5aE

• a break on the monotony that is cyber security research, a sisyphean task that grinds my soul with each hour i spend devoted to it.
• it sounds cool.

It’s probably a mixture of all of the above, honestly, and maybe something more. A lot of it is that I’m garbage at math and objectively need to improve to be more well-rounded.

Regardless, when I saw Ronald Kneusel’s Practical Deep Learning (affiliate link, because C.R.E.A.M) at Books-a-Million during the five minute break I got while my kids were playing with their sticker book and my wife was working on her doctorate, I decided today was the day I started to take a crack at it.

Thus, the study sessions begin. I’ll be writing these up as frequently as I can as an informal transfer of notes and knowledge. Hopefully it will help give back to folks who are half a step behind me, and I hope that the folks who are 1+ steps ahead of me are patient with me.

Session One: Practical Deep Learning

Practical Deep Learning is, thus far, a good book. I’m ~50 pages in on the first day, but a lot of that I skipped because it’s Python primer.

Off the jump, things I’m less thrilled about with the book:

• They seem to focus fairly little on the math. There’s some stuff about matrix multiplication, dot products and a little statistics, but the book seems to be trying to steer away from the math as much as possible. I count that as a negative, but I also acknowledge (as the book does) that the book is a primer on the subject of Deep Learning and ML and thus can’t cover all of the math. I’ve got a linear algebra text book I’ve never so much as opened in my office that will make up for it.
• Python… ugh. The book is Python-based. Those who follow me on Twitter know that I’m on my Rust grind right now and the thought of reading or writing Python code isn’t thrilling for me. I’ll probably write some Python code to figure things out and then figure out ways I can translate into Rust or maybe Go for longer term research. Not docking the book on this either, by the way, I know Python is the go-to for ML research, and maybe I’ll be proven wrong about this later.

The pros, though, thus far outweigh the cons.

• The author writes on heavily technical stuff pretty well. It’s very easy to get very dry on this subject matter, and I’ve been able to read this book without falling asleep, which is high praise considering how brain rotted I am.