It took me about three years to go from deciding I wanted to be a machine learning engineer to landing my first ML job. I picked up plenty of study tricks along the way, and two helped more than the rest: study while you're moving, and split your study time between hard and medium material, hard first. I first gave these tips in 2018 on Machine Learning Applied, the companion show to Machine Learning Guide.
My top walking pad picks for studying on your feet, plus every model I track in a sortable table.
In 2018 I did this at the gym. I loaded my iPad with Stanford and Udacity course videos from YouTube, textbook PDFs and Kindle books, propped it on the treadmill's shelf, and read for the whole workout. Treadmill, elliptical or stair climber, it didn't matter which.
When I tell people this, they usually say they could never focus like that. For me it works the other way around. Sitting still with dense material gives me the jitters, and moving burns them off. With the blood pumping and the endorphins going, I focus better than I do in a chair or in bed. Colleagues have told me the same, and I've seen other people recommend it online.
The bigger reason is that it's two birds, one stone. You want to exercise every day, and you want to study every day. When they're separate, it's easy to do one and skip the other. When they're one session, skipping means missing both, which feels worse than missing one, so you're more likely to do the whole thing.
Today I'd do this on a walking desk: a walking pad under a standing desk. It's the same idea at a walking pace, and it isn't limited to a gym session. I walk at my desk while I code, study, record and edit, around 10 miles a day, which I covered in my setup post.
Give yourself a week to adapt. Read something easy at first, like a novel, to build the habit. After that you'll be able to move to the textbooks without much extra effort.
Split your study time into roughly 25% hard material and 75% medium material, and do the hard part first. Vegetables first, then dessert.
Hard material is the textbooks and the math: Ian Goodfellow's Deep Learning, Artificial Intelligence: A Modern Approach, The Elements of Statistical Learning, and statistics, calculus and linear algebra. It isn't fun to read, and you need it to fully understand machine learning. You'll have to learn it eventually. Rather than front-loading all the fundamentals before you build anything, or putting them off to the end because they're unpleasant, mix a small dose into every day.
Medium material is technical but less grueling: hands-on books and course video series. In 2018 that meant the hands-on TensorFlow book and Stanford's lecture series on YouTube, like CS231n on convolutional neural networks. They take real mental effort, but they're fun, they cover the techniques people are talking about, and they're what gets you building a portfolio.
My routine was an hour of reading a day on the treadmill: the first 20 minutes on hard material, the next 40 on medium. That comes out closer to a third than a quarter, so treat the ratio as a guide. The point is a smaller daily slice of the hard material, done before the rest.
Easy material, like The Singularity Is Nearer or the Great Courses series on consciousness, is inspirational more than educational. It's fun, and you don't need it. Save it for bedtime or a beach vacation.
My machine learning resources page tags its resources as easy, medium or hard and has a Difficulty filter, so you can fill both blocks from it. Goodfellow's Deep Learning, The Elements of Statistical Learning and Mathematics for Machine Learning are tagged hard. Hands-On Machine Learning with Scikit-Learn and PyTorch and Dive into Deep Learning are tagged medium.
Machine Learning Guide is my free audio course on machine learning, from the intuition up through models, neural networks and the math. My newer AI shows are in the free MLG group on Gnothi. You can copy one there, change it to what you want to learn, and Gnothi writes and narrates your version, which you can listen to on that same treadmill.
My top picks for value, budget and premium, motors I've opened to check the brushless claims, and a sortable table of every walking pad I track.