Berkeley CS/Stats v MIT CS

[mathgirl28]

I am interested in Machine Learning and have been admitted to both these programs. I know that Berkeley has MJ, but he is limiting the amount of students he is taking, so working with him might be quite a gamble. There is of course, Martin Wainwright, Peter Bartlett etc. MIT also has some pretty great people as well (Tommi Jakkola etc.). I wondering if anyone has any sort of insight about how to make this decision...

A tentative more specific research area is bayesian nonparametrics... but i like high dimensional data and other topics as well. Again, any input would be great...

[newms]

Congrats on your admits! Were you able to visit the campuses and talk with possible advisors? Fit is always the most important thing, so I would say choose the school you think you would be most comfortable with (in terms of your interests). If you still can't decide based on the more important factors, such as research fit, advisors etc, maybe you could consider which location you would be happier at, since you'll be spending ~5 years there. It's a nice problem to have, congrats again.

[mathgirl28]

I spoke to Peter while I visited Berkeley and he is not leaving. He will be visiting Australia more often. I think he still wants to keep a connection to Berkeley.

Granted, there are quite a number of people from both that i could work with (more than I listed originally). I do like graphical models and nonparametric bayesian research so i think either institution could be a good fit... and that is actually more or less my problem.

Both institutions really impressed me during the preview weekend, and taking out weather as a parameter, one institution didn't really top the other. I am looking to go into academia, so i am really actually trying to base this decision on advisor strength and compatibility rather than weather or superfluous things. While only i can speak to compatibility, i was hoping to get an idea of advisor strength.

[mathgirl28]

Mike said although he wasn't, he has not turned down a strong and eager student then set up meetings for me with people in his group. So I don't think I can count him completely out, just more of a gamble.

Alan Willsky and a couple other people do non parametric bayes work at MIT but it's slightly spread out. Nevertheless you seem to be suggesting that Berkeley might have a slight edge...

Edited March 11, 2012 by mathgirl28

[jjsakurai]

You really can't go wrong with either one. I'd say that Berkeley does have a slight edge especially since the trio of Jordan, Russell and Wainwright seem to collaborate quite a bit. So if you're student with any one of them, you have a good chance to work with the other two. Also there are a ridiculous number of profs at top universities I've seen who've done their postdoc with Jordan/Russell/Wainwright before getting their professorship so if you're looking towards the academia, that is something to consider.

Another thing you should make sure about is what are the chances of getting one of those three as your advisor at Berkeley vs. getting a great advisor (Jaakola, Willsky, Freeman, etc.) at MIT.

[mathgirl28]

It seems though it is more favorable to do a postdoc at Berkeley rather than a phD. The same outcomes hold for phD students, but its not as strong (i dont think...as in you see the same student outcomes for just graduate students for top professors at MIT and Berkeley I believe).

That makes sense because very strong graduate students are more likely to get selected for postdoc positions with these three, and generally the name recognition of having them as their advisor is primarily a resume boost, as they were overall very strong anyway (and probably would have landed one of those postions)...