Andrew Ng updates his Machine Learning course (opens in new tab)

All videos of all courses in Coursera are free. You can watch them fully without providing your credit card info.

There are two types of courses in Coursera- free and paid.

In case of the paid courses, you can go to the course and navigate to the "Buy Subscription" page and click on "audit the course". You can watch all the videos for free, but you don't get access to quizzes and programming assignments (you never know what a web search will turn up ;)) ⊕. You do not get a certificate by completing a course or completing all courses of a "Specialization".

In the case of a free course, you get access to all the videos, quizzes, and assignments. You don't get any kind of certificate. Instead of going to subscription page, you can just click "Enroll" and choose the no certification option.

There are some great courses in the free tier (videos + assignments, no certs) as well. Dan Boneh's Cryptography and Grossman's Programming Languages A, B, C come to mind. Also Model Thinking by Scott Page.

There were some great discussions on HN in the past. [0][1][2]

⊕ There are courses where duplicates of paid assignments and quizzes are provided under "Practice Assignment" as opposed to "Graded Assignment". Like Martin Odersky's Functional Programming Principles in Scala MOOC.

[0]: https://news.ycombinator.com/item?id=25245125

[1]: https://news.ycombinator.com/item?id=16745042

[2]: https://news.ycombinator.com/item?id=22826722

IIRC you need to pay if you want your assignments to be (auto-)graded.

That link says "Enroll for Free" and no audit button. Maybe it's because I'm not logged in?

Coursera has a monthly $45 fee for the whole specialization. But, the specializations also include a 7 day trial. You get access to all course material and assignments and all courses. Back in my college days(2-3 years ago), I would start a specialization and blaze through in 6-7 days. Money saved and time well spent. Ofc now that I am working, it's not going to be that easy.

You can also purchase an yearly Coursera subscription for $299 or $399 and get access to all the specializations/projects on Coursera for one year.

I still remember a talk by a woman from Google at a fairly long ago now O'Reilly conference (R.I.P). Part of what she discussed was Research AI vs. Applied AI. The gist of it was that a lot of the things in university course, graduate programs, etc. are tilted towards Research AI and you can get away without a lot of that stuff by using pre-built tooling for practical machine learning applications.

Of course, you want to have some understanding of what's going on under the covers but, for a lot of people, starting from first principles is quite hard and isn't really necessary.

Is the knowing only Algebra I enough for this course?

Can you explain a little more about how and why this was practical?

Do you thing now days Deep learning does not requires much math? If yes, to what extend of knowing math is enough to be truly good deep learning specialist? By deep learning specialist I mean the person who is building a commercial software that uses deep learning but not tools for deep learning.

There's a lot of ML happening outside of big corporations, which you can confirm by just searching 'machine learning' on any job site. While it's true that often you can use ready-made ML solutions, you often will benefit from additional knowledge for improving or adjusting them for your company's specific problem and while interviewing you will often be asked the kind of questions those courses cover.

You can get almost unlimited GPU time on Google Colab for $50 a month. I don't know why or how they pay for this, but it does bring "real research" into the reach of individuals.

> You could use ML in your job/company but then you dont need this course, you just use a ML product.

ML product?

Or maybe people want to understand what's going on under the hood of the ML products they use?

How about joining FAANG as SWE, and then internal transfer?

I highly recommend https://course.fast.ai/. It's much more top down: in the first lesson or two, you train a NN image classifier, rather than starting with first principles and linear algebra. I found this structure to be more motivating and effective.

I took these courses from Georgia Tech via OMSCS but they are also on udacity.

https://omscs.gatech.edu/cs-7641-machine-learning

https://omscs.gatech.edu/cs-7642-reinforcement-learning (I took this before ML but its supposed to come after. There is some overlap. Probably my favorite graduate course.)

https://omscs.gatech.edu/cs-7646-machine-learning-trading (IMO not amazing)

Much more basic (took this before OMSCS):

https://www.udacity.com/course/intro-to-machine-learning--ud...

I'm sure there are many more.

Not a full course I'd say, but I've used this one to learn the math behind deep neural networks and code my own from scratch in elixir and C:

http://neuralnetworksanddeeplearning.com/

"Learning from Data" is outstanding: https://work.caltech.edu/telecourse.html

It's a recorded version of a real Caltech undergrad course, and it's focused on understanding the math behind these algorithms, not just applying black-box ML libraries.

It's much less practical, but I feel like it teaches you more.

Depends, how much linear algebra, probability and python do you know?

If you remember highschool AP math, you're good. Otherwise, check out ISLR for a faster intro, capped by your ability to read

Definitely. It starts from first principles, linear algebra, and goes from there. It's an amazing course.

Before Python (Numpy/Scipy) really came into its own (which didn't quite happen until early 2010s), Matlab was among the easiest-to-use scripting languages for writing scientific computing programs. I was in university (Bachelor's + Masters) from 2007-2012 and learned Matlab extensively in my numerical computing classes (I was a Physics major, for what that's worth). When you're ready to run your scientific computing codes on a supercomputer cluster, you'd usually rewrite it in C++ or Fortran (my research group used the latter), but to develop and debug at a small scale, you'd usually use Matlab, although the younger people like me might use Numpy, and there were one or two people who used Mathematica even for numerical computing (as opposed to symbolic algebra, which everybody used Mathematica for).

It was around the time I was in university that Python really matured for numerical computing, but professors (as opposed to grad students) were likely to be already familiar with Matlab, so there wasn't much reason for them to learn Python. Andrew Ng was already a mid-career researcher when he made his course, which was probably based on older materials (I also learned basic neural networks in my numerical computing class in 2008), so it made sense for him to continue to use Matlab, especially because Octave exists as an open-source reimplementation of the basic functionality.

These days, you wouldn't use anything else but Python for ML, at least until you really productionize the implementation at a large scale, at which case you might rewrite in C++ or Rust (I don't know if they even bother rewriting these days when most of the computation happens in GPUs or TPUs). And it's my understanding, although I'm not really too familiar these days, that Matlab has mostly pivoted into providing a toolbox of all sorts of esoteric numerical methods for engineering-related tasks like finite element analysis, as well as hardware simulation (using Simulink).

10 bucks says it's python

OTOH, the time I spent learning Octave/Matlab for Andrew Ng's course was 100% wasted time, because I've never used it again in the 10+ years since I took the class, whereas time spent learning Python would've been useful to me in myriad other ways.

The programming assignments in the original course were mostly useless. They provided you with a template with 90% of the problem solved, and you just had to enter an equation to solve the problem.

Yes - learn yet another language in the Tower of Babel of languages that will be useless after this course. There's a reason he switched to Python.

I mean python has issues (indentation errors isn't one I would list) but that's besides the point isn't it. The Lingua Franca for ML is currently python. Teaching octave, when most things they search for will be python just seems unnecessarily stubborn. Some day it might be Julia but we aren't there yet.

Python is an easy language for beginners, and with numpy, it has excellent support for linear algebra.

But it's the language dujour for ML and in particular Deep Learning so theres no point doing it in any other language.