Re:Stanford Machine Learning course
by physcab on Tuesday February 03, 2009 @12:48PM
(#464316)
I'm in a Machine Learning class right now and the math is absurd. Our professor goes over general theory in class and references some practical applications. For homework we have to prove the theories and then create MATLAB programs that create the figures in our textbook (Bishop). It's pretty dense and time consuming.
However, I come from a Physics background, so the math is just different, and may in fact be easier. However, I've been referencing my Mathematical Methods for Physicists book (Boas) pretty often to brush up on some linear algebra techniques.
Lastly, its been said quite a bit in these forums, but if you are just interested in implementing this stuff, definitely check out Collective Intelligence by Oreilly. I use that book in tandem to bring some of the high-level concepts back down to earth. :)
Re:Stanford Machine Learning course
by FraaJad on Tuesday February 03, 2009 @09:27AM
(#464016)
I read some of the ML lectures notes from OCW some time back. It was very math heavy. Is that an MIT thing or is that how ML is taught in all universities?
Re:Stanford Machine Learning course
by jerf on Tuesday February 03, 2009 @12:51PM
(#464318)
I can not conceive of how one would have a "non-math heavy" machine learning course. Machine learning is basically a branch of applied statistics, at least in its current form. None of the algorithms are "fire and forget", such that you can use them effectively without deeply understanding them, and one of the first things you'll cover in your Machine Learning course is why it is unlikely that will even happen (the No Free Lunch theorem, often over-applied in internet debates but still a very, very important result).
(I mean, sure, you can try to fire-and-forget, but for the most part you'll either fail with no clue why, or fail to get very good results. Decision trees might be a moderate exception, although even then you really ought to understand how they are constructed and the implications of various choices you can make.)
Re:Stanford Machine Learning course
by kurtosis on Tuesday February 03, 2009 @02:32PM
(#464485)
As a counterpoint, I seems to me that the ML field is too heavy on theory. A lot of the theoretical papers are proposed with weak or no experimental tests. There are very few people doing thorough experimental work. What I have in mind is something like Chen and Goodman's comparison of different language models - this is a great paper and I'd love to see more like it.
The obvious example of this is boosting where the theory lagged far behind the experiment - everyone wanted to invent a theory to explain why boosting was so effective.
However, I come from a Physics background, so the math is just different, and may in fact be easier. However, I've been referencing my Mathematical Methods for Physicists book (Boas) pretty often to brush up on some linear algebra techniques.
Lastly, its been said quite a bit in these forums, but if you are just interested in implementing this stuff, definitely check out Collective Intelligence by Oreilly. I use that book in tandem to bring some of the high-level concepts back down to earth. :)
Good luck
The midterm for CS 229 was probably the hardest exam I took in college.
(I mean, sure, you can try to fire-and-forget, but for the most part you'll either fail with no clue why, or fail to get very good results. Decision trees might be a moderate exception, although even then you really ought to understand how they are constructed and the implications of various choices you can make.)
The obvious example of this is boosting where the theory lagged far behind the experiment - everyone wanted to invent a theory to explain why boosting was so effective.
hey industrialists out there - publish!