Learning ML with no background

Notes and thoughts on learning machine learning. This guide is dedicated to the one and only Adnaan Sachidanandan.

The classic Andrew Ng course on Coursera is a good place to start. You don't have to finish the whole thing but it's a good introduction
Andrej Kaparthy's hacker guide to neural nets is very helpful for anyone with programming background. Go through and actually implement it yourself without copy-pasting, and get a neural net to classify something easy (i did RBG colors -> "blue","green")
Stanford cs231n and cs224d are great, and they post lectures online. Do the projects/homework too! It's probably the most useful part since you learn tensorflow.
Congrats, at this point you'll have a good enough background to get started on whatever project you want. Build a bunch of things and you'll learn a bunch of things.

I tried a lot of different starting points, but my personal favorite has been Neural Networks and Deep Learning by Michael Nielsen. It's written very well, and someone with basic matrix and calculus knowledge can get up and running with machine learning quickly.
If you like videos, Siraj Raval's YouTube channel has a lot of intuitive explanations of the math behind some of the major concepts and also describes some of the newer research coming out of the field.
To reiterate what @kvfrans said, Stanford's CS231n and CS224d are really good places to get your feet wet with building out networks and turning some of the math into code.
FastAI is a great starting point for anyone with some programming experience. I've watched some of the lectures and they are really helpful, especially with getting you off the ground with deep models.
Build lots of things. I started learning by modifying toy code I found online. I then started building projects that interested me using any resources I could find. If you're lost for inspiration, check out Kaggle, there are plenty of interesting problems people are trying to solve and occasionally companies post competitions (with fairly large payouts) for those who can build the best models.

Andrew Ng's new Deep Learning course. Haven't gone through it, but looks good.
3Blue1Brown's YouTube playlist does a good job of animating and explaining the math behind many of the major concepts in deep learning.
It doesn't really matter if you use Tensorflow or Pytorch or anything, just pick one to start with. Also, implementing backprop by hand is a good exercise you can figure out what's happening. Don't actually use it for anything big though.
When working on projects don't be afraid to steal parts of code from other git repos / stackoverflow. Our philosophy: do whatever it takes to make your project work. If you work on interesting enough things, the experience will come naturally.

kvfrans/learn