Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Foundations Machine Learning

bloomberg.github.io

Post reply on HN
Print view View original

11–20 of 61 posts

Re: Foundations Machine Learning

#11

Post by ghaff »

Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

Some intuition about minima (gradual descent), linear regression, etc. will get you to a certain level--which admittedly may be a bit hard if you've never had the math. But as others have suggested, you probably need to equivalent of a few semesters of calculus and linear algebra (which I never took in a formal way though I did some variants of pre-MatLab).

Re: Foundations Machine Learning

#12

Post by Jasonseah »

backpropaganda wrote: post #7
Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

What is your math experience/background? In any case, for the average case, this course is probably the best for those who want a slightly deeper understanding of machine learning than offered by Andrew Ng's courses: https://work.caltech.edu/lectures.html

can say at a very basic level ... I don't really focus on the study at my early age. I still can understand the equation but not really know why and how to implement it, and come to a very complex situation I can't make sense on the calculation ... Every calculation that Andrew Ng teach are new to me even for the matrix, vector, and sum over are new to me. Even tho Andrew Ng did say "if you don't understand math is ok." but I still love to understand math for my machine learning career, that's why I seek for help.

Re: Foundations Machine Learning

#13

Post by cbHXBY1D »

hackermailman wrote: post #5

If you know basic undergrad probability/statistics/linear algebra (matrices, vectors, eigenvalues) you can do CMU's graduate course in ML that is intended to prepare PhD students to understand research papers in the field (Includes recorded lectures) https://sites.google.com/site/10715advancedmlintro2017f/lect... CMU also has an 'Applied Machine Learning' undergrad course that is paywalled fully unfortunately, but th…

And what about after that? In my experience at least, I can highly recommend Stanford's EE364a Convex Optimization. http://web.stanford.edu/class/ee364a/

When I was an undergrad at Berkeley, one particularly well known research lab would only look at your resume if you took that course online. Warning: that class is not for the faint of heart. And be good at linear algebra.

Re: Foundations Machine Learning

#14

Post by Jasonseah »

ghaff wrote: post #11
Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

Some intuition about minima (gradual descent), linear regression, etc. will get you to a certain level--which admittedly may be a bit hard if you've never had the math. But as others have suggested, you probably need to equivalent of a few semesters of calculus and linear algebra (which I never took in a formal way though I did some variants of pre-MatLab).

I did try to play around with Octave / MatLab to play with the calculation but not in a very comfortable way ...

Re: Foundations Machine Learning

#15

Post by Jasonseah »

siddboots wrote: post #8
Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

If you want to keep up with the maths, the main prerequisites for more rigorous ML is typically undergrad-level calculus, linear algebra, and bit of basic probability and statistics. For calculus, google "MIT 18.01", "MIT 18.02", (and "MIT 18.03" if you like), which are all freely available on youtube. You should be comfortable with single-variable calculus, and at least familiar with multi-variable techniques. For l…

Thx! will look into it!

Re: Foundations Machine Learning

#16

Post by Jasonseah »

hackermailman wrote: post #9
Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

Depends what your math level is. Start with https://www.expii.com/ which is Po-Shen Loh's site and see where you are. If you can do the algebra and basic calculus problem sets there you could also try these specific ML background books https://sites.google.com/site/10715advancedmlintro2017f/read... Expii is a great resource because it determines from your success solving problems if you need more practice to get a co…

Thanks for the suggestion! Help a lot for what I am struggling now.

Re: Foundations Machine Learning

#17

Post by colmvp »

hackermailman wrote: post #5

If you know basic undergrad probability/statistics/linear algebra (matrices, vectors, eigenvalues) you can do CMU's graduate course in ML that is intended to prepare PhD students to understand research papers in the field (Includes recorded lectures) https://sites.google.com/site/10715advancedmlintro2017f/lect... CMU also has an 'Applied Machine Learning' undergrad course that is paywalled fully unfortunately, but th…

Alas, it seems that they don't have videos for all their lectures?

Re: Foundations Machine Learning

#18

Post by colmvp »

siddboots wrote: post #8
Jasonseah wrote: post #6

I am new in machine learning and have difficulty to understand the equation. I finish 2 courses from Andrew Machine Learning Class from Coursera, I understand the flow, concept and knew how to write those equations/algorithm but it always bugs me that I don't understand those equations. Do you guys have any suggestion that where should I start for learning those equations/maths online?

If you want to keep up with the maths, the main prerequisites for more rigorous ML is typically undergrad-level calculus, linear algebra, and bit of basic probability and statistics. For calculus, google "MIT 18.01", "MIT 18.02", (and "MIT 18.03" if you like), which are all freely available on youtube. You should be comfortable with single-variable calculus, and at least familiar with multi-variable techniques. For l…

While I think 3b1b video's are great for developing a better understanding of the purpose of certain concepts, I don't think there's any way around ignore manually doing the problems with say Strangs Linear Algebra book since that's where you spend tim trying to apply these concepts to problems.

Re: Foundations Machine Learning

#19

Post by hackermailman »

cbHXBY1D wrote: post #13
hackermailman wrote: post #5

If you know basic undergrad probability/statistics/linear algebra (matrices, vectors, eigenvalues) you can do CMU's graduate course in ML that is intended to prepare PhD students to understand research papers in the field (Includes recorded lectures) https://sites.google.com/site/10715advancedmlintro2017f/lect... CMU also has an 'Applied Machine Learning' undergrad course that is paywalled fully unfortunately, but th…

And what about after that? In my experience at least, I can highly recommend Stanford's EE364a Convex Optimization. http://web.stanford.edu/class/ee364a/ When I was an undergrad at Berkeley, one particularly well known research lab would only look at your resume if you took that course online. Warning: that class is not for the faint of heart. And be good at linear algebra.

CMU also has a convex optimization course w/open recorded lectures on youtube http://www.stat.cmu.edu/~ryantibs/convexopt/

None of these classes are for the feint of heart if you're actually enrolled in them and operating under crazy deadlines with thousands of dollars on the line. This is clearly for our casual interest and hacking around with ML research hobby.

Re: Foundations Machine Learning

#20

Post by siddboots »

colmvp wrote: post #18
siddboots wrote: post #8

Earlier quoted context omitted.

If you want to keep up with the maths, the main prerequisites for more rigorous ML is typically undergrad-level calculus, linear algebra, and bit of basic probability and statistics. For calculus, google "MIT 18.01", "MIT 18.02", (and "MIT 18.03" if you like), which are all freely available on youtube. You should be comfortable with single-variable calculus, and at least familiar with multi-variable techniques. For l…

While I think 3b1b video's are great for developing a better understanding of the purpose of certain concepts, I don't think there's any way around ignore manually doing the problems with say Strangs Linear Algebra book since that's where you spend tim trying to apply these concepts to problems.

Agreed. Nothing beats practice when it comes to math. I know that I spend far too much time watching course videos, and no where near enough working through problem sets.
Post reply on HN