The Big Dum-Dum’s Guide to Learning how to Deep Learning…. Learn Deep.. ROBOTS!
Or how I stopped caring and learned to learn to love math, or at least tolerate it… kind of.
Do you know what this means?
It’s essentially how to tweak the knobs on an algorithm so it can get closer to predicting the correct classification of a data point. Aka learn…. deeply. Four weeks ago I could not have cared less. Scratch that. Math notation would have made me angry. In fact if you had showed it to me, I would have punched you.
I thought I was done with math when I got a C in pre-calculus then dropped out of college a semester later. Fast forward 15 years later and here I am kind of starting to getting to understanding the beginning of knowing a little bit about calculus….
Why the masochism? I enrolled in the Deep Learning Foundations NanoDegree at Udacity about three weeks ago, kind of on a whim. Now I’m studying my ass off every day to understand the calculus and the concepts behind deep learning. Two weeks in and it’s already been one of the most rewarding learning experiences of my life.
I’m writing this primarily for my O.G. DLFND homies (whatup?!?) in the first cohort and hopefully for the other cohorts to come. Take it from someone who started with less math skills than a freshmen in high school, if I can get this far, you can learn it… probably.
The following may be obvious to a lot of people but it wasn’t for me. Here’s what’s worked for me, so far:
- Go through every lesson and write down what you do and don’t understand. Try to make connections from formulas and models to written out definitions.
- When you begin to understand a new concept go back and rework all the lessons before. Example: Once Gradient Descent starts to make sense, start back in lesson 1 of Intro to Neural Networks. And yes, hit reset quiz and redo that code exercise.
- When you can’t seem to grasp a new concept go back even further and rework. Example: Once your mind has melted from trying to grok Back Propagation go back to Regression Models and work back up to the mind melting. And don’t even think about hitting next without hitting reset quiz and resubmitting the code exercise!
- Focus, for at least 2 hours in a row. With no distractions. No youtube, no twitter, no facebook (not even to read about pytorch), No Medium (get outta here!). And shout out to my DL homies in the DLFND #beginners Slack channel but you got to shut that shit down while you’re studying. My goal, every day is to spend 2 hours in the DLFND coursework and 1 hour on Khan Academy. Some days I make that goal some days not, but I’m at it nearly every single day.
- Khan Academy is your friend. Start with Multi-Variable Calculus. If that’s too hard (like it is for me) back propagate to Precal or Algebra.
- Cherish every small victory. Every concept that I grok just a little more is getting me one step closer to my harem of sexrobo… I mean army of killer rob.. I mean having a profound and positive impact on humanity. Let’s be real, to a newbie, Deep Learning and AI is way overwhelming and hard to learn. But if you look at it like a bunch of baby steps it’s very doable (props to ’What about Bob?’).
Did you catch the lame back propagation reference in Step 5? I learn how computers learn?!? Queue the overused Tim and Eric meme!
If you’re reading this more than a month after I wrote it and you’re curious, leave a comment to see how my studies are going. I could use the accountability.
Oh shit! Gotta go. My first project is due in one week!
See you in the Singularity!
-Adan Gabriel Gutierrez