Any machine learning folks on this forum?

For the discussion of math. Duh.

Moderators: gmalivuk, Moderators General, Prelates

User avatar
MartianInvader
Posts: 771
Joined: Sat Oct 27, 2007 5:51 pm UTC

Any machine learning folks on this forum?

Postby MartianInvader » Fri May 19, 2017 11:39 pm UTC

Hi guys, I'm wondering if someone could help me with the following question:

Of all the algorithms and models out there labeled as "machine learning", which has the broadest applications? Which is most widely used, either directly in applications or as influence on other models? And what other algorithms and models is it related to?

The question is obviously subject to opinion - there's no rigorously "correct" answer, but I'm curious what the top contenders would be.

My motivation is that I have a math/programming background and I'm playing around with some really vague notions of machine learning, but I know very little about it - I've seen a couple variants of simple neural nets, but that's about it. I don't really have time these days to learn enough of the subject to answer the question above, but it would be nice to have a good model I understand and can play around with.
Let's have a fervent argument, mostly over semantics, where we all claim the burden of proof is on the other side!

>-)
Posts: 509
Joined: Tue Apr 24, 2012 1:10 am UTC

Re: Any machine learning folks on this forum?

Postby >-) » Sat May 20, 2017 3:16 pm UTC

linear regression models are pretty fundamental and widely used -- in fact, my school has an entire semester long course on various forms of linear regression. and i would bet that in terms of "number of times applied", this would be the winner

two other good ones are decision trees + their variants, and SVMs.
although i've never participated, it seems that a large portion of data-analysis / ML competitions are won by some form of gradient boosting decision trees. SVMs are a pretty popular tool too, and there is a huge amount of theory built up on them, if you are more of a theory person

finally for any structured high dimensional data (images, video, audio), deep neural networks seem to perform better than almost everything else, and i would expect them to remain dominant for quite a while. in terms of "number of bytes of data processed", these probably come out ahead

Newt
Posts: 57
Joined: Sun Jul 24, 2011 2:33 pm UTC

Re: Any machine learning folks on this forum?

Postby Newt » Mon May 22, 2017 6:44 pm UTC

From an industry perspective, I wouldn't invest much time looking into SVMs; neural networks have largely displaced it in most application areas. Tree-based methods are popular for general business problems (e.g., https://www.kaggle.com/c/otto-group-pro ... -challenge); linear models generally lag behind them in terms of predictive power, although they've been shown to be a contender when you have high volume, high dimensionality data. Linear models are also competitive against more complicated neural network approaches on some natural language processing tasks.

Matrix factorization methods have some applications in recommendation engines/NLP.

Neural networks are dominant in the areas most people traditionally think of as AI (I recommend Bengio's textbook on the subject if you're just getting started)

User avatar
MartianInvader
Posts: 771
Joined: Sat Oct 27, 2007 5:51 pm UTC

Re: Any machine learning folks on this forum?

Postby MartianInvader » Mon May 22, 2017 8:07 pm UTC

Thanks guys! Seems like neural networks might be a good place to start for me.
Let's have a fervent argument, mostly over semantics, where we all claim the burden of proof is on the other side!


Return to “Mathematics”

Who is online

Users browsing this forum: Google Feedfetcher and 13 guests