Hacker News new | past | comments | ask | show | jobs | submit login
Course: Mathematics for machine learning (coursera.org)
106 points by nafizh on April 9, 2018 | hide | past | favorite | 10 comments

This course is from Imperial College in London. It's typically ranked as one of the top 3 universities from the UK. Finally it has joined the ranks of doing a coursera course. Other noticeable absentees from coursera are Oxford and Cambridge. I know they do provide a few of their own though open courses.

I guess universities shouldn't be forced to give open courses, but it is a fantastic thing that US showed was possible and I hope that more UK universities follow suit.

Oxford actually has one course on EdX: https://www.edx.org/school/oxfordx

My understanding is that, like many MOOCs, it is largely department/faculty driven rather than a strategic or philanthropic move by the university overall.

I've been looking for something like this to brush up/add to my math knowledge; can anyone recommend this course or would you recommend some other way?

Many ML methods require solid knowledge of probability and statistics. Strangely this course does not cover that.

My impression from reading about a few ML techniques (such as Neural Nets or Support Vector Machines) is that this class of ML algorithms relies heavily on regression techniques that are, at their core, nonlinear optimization problems. This means finding local maxima and minima for systems of nonlinear equations, either analytically or through gradient descent. Either way, you're going to need to deal with a system of partial derivatives, which means you need to understand vector calculus and linear algebra. If that's the focus of this course, then I'd say it does make sense.

I understand what I described above is a subset of ML, and I generally do agree with you that a solid background in probability and stats is important for people who plan to do a lot of ML.

Note - another possible objection here is that all "STEM" fields require calculus through differential equations and linear algebra. That's pretty much the common thread for most majors generally grouped together as STEM fields. So calling this "mathematics for machine learning" could be a little strange. If we're going to call vector calc and linear algebra "mathematics for ML", why aren't we calling it "math for physics", or "math for engineering"... I think that part of what is going on is that ML has gotten very popular, and people are starting to ask what the essential math background is, and are discovering that it's, well, pretty much the two year science and engineering track calculus you'd take at most universities.

Funny, you used the term “regression”. People without stats 101 background will not know what you’re talking about.

I agree - it looks equivalent to perhaps 1st year mathematics in Australian university or 2nd year 1st semester if I'm being really generous.

This definitely isn't sufficient for machine learning, but it is a start.

It is good for understanding many ML algorithms and how they work internally. So in my opinion it's enough.

Ah, I should've figured that out myself from looking at the contents. I might try it out and combine it with some reading of my old statistics book or some other means. Thanks!

Link is dead


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact