Stop focusing on MOOCs and youtube videos and study textbooks. Do exercises. Treat it like academic studying, and you'll end up with a decent education. It's important, because it's often easier to make a thing work okay than to understand why it works, so you'll get false confidence working through a tutorial. But then you want to apply that to something else and it doesn't work quite right, you won't know why it doesn't work and how to fix it.
Get some textbook suggestions and make a minimum of reading 5-10 pages per day. In about a month or two, you're done with a 300 page book. Repeat that for a few years and you're an expert. Once you have the foundations, read papers too, but don't skip straight trying to using AlphaZero to solve a curve fitting problem.
> study textbooks. Do exercises. Treat it like academic studying
This. Highly recommend Russel & Norvig [1] for high-level intuition and motivation. Then Bishop's "Pattern Recognition and Machine Learning" [2] and Koller's PGM book [3] for the fundamentals.
Avoid MOOCs, but there are useful lecture videos, e.g. Hugo Larochelle on belief propagation [4].
FWIW this is coming from a mechanical engineer by training, but self-taught programmer and AI researcher. I've been working in industry as an AI research engineer for ~6 years.
Oof those are all dense reads for a new comer... For a first dip into the waters I usually suggest Introduction to Statistical Learning. Then from there move into PRML or ESL.
Were you first introduced to core ML through Bishop? +1 for a solid reading list.
PGMs were in fashion in 2012, but by 2014 when Deep Learning had become all the rage, I think PGMs almost disappeared from the picture. Do people even remember PGMs exist now in 2019?
You're not going to learn much of anything by just reading a ML/AI book. Well you gotta be pretty good at math to understand everything, my suggestion is to enroll on some ML/stats course and start working on basics. It's long and hard road, and if you're not comfortable with math in general i'd reconsider investing too much time on it. I'm doing it yet i'd much rather do software engineering. Much more practical without math to solve.
So do exercises, spend time digesting and trying to explain things to others. If you feel it's hard, well you are correct. Get comfortable feeling that way. Hopefully theres light at the end of the tunnel. Dont buy into the hype. Know the basics
Edit: so didnt see op said exactly this. My bad, new year and all.
I'd need a lot more remedial mathematics before I could get any mileage out of those ML/AI book recommendations. I haven't touched mathematics in a strict sense since my senior year of high school Calculus (I majored in the humanities), so I think you're absolutely right. People like me would need to spend a lot more time learning things like discrete math and statistics before moving up to these books.
This is exactly my experience. I started up with blogs and sites. Though they were good when I just started out but after a point, I failed to make a coherent, systematic and deeper understanding of the topic. I felt like I am getting knowledge in bits and pieces which weren't creating a complete and package.
Finally I started studying serious books in a disciplined manner. I wish I should have done this earlier.
> Stop focusing on MOOCs and youtube videos and study textbooks.
I'd be ecstatic if I never again see a comment about how folks suddenly and completely understand a class they failed years ago after watching a 3blue1brown video.
I think moocs are usefull to jump start you but after that you need to do more. I studied CS and did find the deeplearning.ai course helpful in getting started. Without it much of the ML content was not easy to grokk. But now that I"ve gone through that I get the gist of what papers are talking about. After you do a MOOC though you have to continue with doing real work as in exercises, Kaggle competitions and just playing with things.
I agree that MOOCs, particularly in areas such as math and physics, are woefully insufficient.
Nevertheless, videos and MOOCs still have their place, because they help you formulate the questions you need to answer. They help you know what unknowns you don't know.
My experience has been different. "Academic Studying" can be hard with a full time job, and there is often a wide gap between what you learn in books and what you do at the job.
For purely professional purposes, it's probably a better idea to take a non-academic approach.
Yeah, it's super hard with a full time job. That's why I recommend setting a low X pages per day target. Making sure to make just a few pages of progress per day really adds up quickly. You likely won't progress as fast as if you were a full time student, but that's okay. It still adds up fast enough.
A minimum of 5-10 pages a day for 6-12 months seems pretty reasonable for a career change into a competitive field.
Thanks for the recommendation. The Murphy's book looks solid, better than half the books above. I personally own the ISLR(Hastie) and learning from data(Yaser). Both are beginner friendly. What were the Math requirements for the classes that you took at Uni that used Murphy's book? And how would one go about acquiring those Math/Stats/Prob skills in order to work through a tome like Murphy?Thanks
Get some textbook suggestions and make a minimum of reading 5-10 pages per day. In about a month or two, you're done with a 300 page book. Repeat that for a few years and you're an expert. Once you have the foundations, read papers too, but don't skip straight trying to using AlphaZero to solve a curve fitting problem.