DMX

TTIC is indeed very well regarded. Probably not top 10. I'd say it is comfortably in the top 20 though.

You will not face any issues if you decide to go to USC, especially since you are not getting aid at NYU.

Hey, a few things. - Unless you publish at a top ML conference by the time you apply, I am sorry to say that your chances don't look good. Excellent research records can certainly overcome low undergrad GPA but your two papers are at venues that mainstream ML people don't really care about (of course, this has nothing to do with the quality of your publication) - Given your background in biomedical ML, try getting in touch with faculty doing biomedical ML and inquiring about your chances. Most top CS programs have at least one professor working in biomedical ML. - You might have a slim shot at programs in the top 10~20 range (which are still excellent schools, btw!), if there is good research fit. But I would apply to a wide range of programs, focusing on those in the 25-50 range. Best of luck!

You are probably on the waitlist if you haven't heard back from Berkeley.

Same. In the letter it says only 1 or 2 people are typically accepted off the waitlist.

I was very lucky to have been accepted into a few schools, and now I am choosing between Harvard and CMU. My interest is in machine learning. I have to make a decision without visiting (international student). I know CMU has a much better CS reputation than Harvard, but I am drawn to Harvard for several reasons: - Had a chance to interview with the professor who would be my advisor, and we got on really really well. - Harvard CS is getting a huge boost in funding from Steve Ballmer (http://www.seas.harvard.edu/news/2014/11/ballmer-to-support-computer-science-expansion) - Boston >> Pittsburgh - Better overall brand (understand that one shouldn't look at this for PhD programs, but it doesn't hurt) CMU seems like the obvious choice. It has a whole department dedicated to my field. CMU's CS is top 4. My CS friends think I would be crazy to turn down CMU for a much lower-ranked school... but I can't shake off the feeling that I would be happier at Havard Can anyone offer advice?

Oh, for personal reasons I am only looking for US grad programs...

Hi. Does anyone know which schools have late deadlines (i.e. Feb or later) for computer science? I ended up applying only to "reach" schools and I am slightly freaking out. So I am looking to apply to a few more schools now...

realistically, i don't think you have a shot at any top-50 phd programs. i would suggest getting into a decent MS program (since it seems like you have tuition money saved up), doing really well, and then applying to phd afterwards. best of luck

go to stanford and don't look back.

Since you already know R, it may help to learn SAS. For SQL, check out a couple of SQL-videos in this MOOC: https://class2go.stanford.edu/db/Winter2013/preview/ If you know programming you should be able to go from beginner-->intermediate SQL programmer in a week. Any reason why you specifically want to work for the government (in which case SAS would be important)?

Regarding number (2), I am in somewhat of a similar situation. How much will it hurt? Have you had cases of admitting students with non-official research (through course projects, employment)? If so, what made them stand out (grades obviously, but anything else)? Thanks!

Nice GRE subject score. However, without publications, international competitions, or recommendations from well-known professors (which may be hard given your international background), you will most likely be out of the running for the top schools (Princeton, Harvard, NYU etc.). Math PhD pool is very, very deep. Of course, you can still apply to them, but I would set your sights a little lower. University of Washington may be in the ballpark for you--have a friend with similar stats who went there.

if you've taken analysis you should have no problem jumping into upper level probability/inference classes.

I can see the main factor being how well-known your university is. (i.e. is it one of McGill, Toronto, Waterloo etc., which everyone has heard of, or is it a place lower down on the list)? Keep taking math classes, though anything beyond analysis is overkill. Instead of topology/complex analysis I would focus on more advanced stats courses (regression, nonparametric stats, inference, and the like). If you are able to get a publication or two under your belt in the upcoming year, I imagine you will be competitive at all programs. Best of luck!

i'm in a similar boat (interested in ML, but not sure if stat/cs/or is right for me) and here's my take on it. CS: you are gonna have a hard time cracking top CS departments. While publications will certainly add to your application, top CS programs are so competitive that they expect CS-related publications (i have a friend who wanted to transition from physics to CS and had publications in top physics journals. but she was unable to crack top 15 and had to settle for 'lesser' schools (which are still good of course!)). OR/applied math: similar story here. OR is extremely competitive (though not as much as CS), and there aren't many applied math departments doing (serious) machine learning. stat: better chance here. applicant pool is shallower. i think you will have a good shot at a decent number of top 15 places. econ: your best bet. i know econ is certainly competitive, but the acceptance % is deflated by a lot of econ grads who haven't even taken multivariable calculus. as you are probably aware, key to cracking CS/OR departments in relevant publications. if you get some publications in relevant journals/conferences by 2015/16 you may have a good shot. good luckl!

Stats will be more competitive than Data Science. I would say you have a good shot at the Data Science programs (which schools have PhD's in data science?). For stats, I would look a little lower down the list as well (though you should have good shots at Penn/Rutgers). Difference in threshold for PhD/MS is also HUGE. My evaluation is for PhD programs. For MS, I think you will be competitive at most places.

I strongly disagree with the recommendation for NC State. This is a Master's degree. As such, brand matters a lot, since it seems like you will go into industry. Look into: For non-online programs - Columbia (MS program starting this year) - NYU (MS program started last year) - Stats/CS degree from a strong machine learning school (Stanford comes to mind) For on-line - Berkeley - Northwestern

I would say it depends on your transcript. If you have a decent number of courses after calc/linear algebra (analysis, topology, number theory, etc.), and you did well on them, then no need to send it. If you have only done up to calc/linear algebra, maybe you should send it to differentiate yourself. 730 isn't half-bad by the way, especially when you consider that most people taking the test are Math PhD applicants.

What are the bare minimum pre-reqs for CS PhD programs? I have a physics/math/stats background, so I've taken my share of upper level math classes, but very few CS classes (I've taken an algorithms class and a machine learning class, but that's about it. Picked up programming on my own but I would say I am a fairly novice programmer). I have a chance to take some grad-level CS classes at a nearby university, and would appreciate advice on what I should take. My interest is in (surprise surprise) machine learning.

Would suggest Math (but of course, take some stats courses. measure theoretic probability, inference, etc.)

dude, you will get in everywhere. Add Stanford to the list if you are interested in ML.

If MS will be to get a job, then MS in data analytics will most likely be more useful. In MS stats you will be learning about things like Cramer-Rao lower bound, sufficiency, power of a test, BLUE, etc. These are beautiful things to learn, but not much practical use. With an MS in Data Analytics, you will (hopefully) be learning about machine learning, big data etc., which are incredible skills (especially nowadays) for the job market. Of course, you can learn the aforementioned skills in a Stats program, but that won't be the "meat" of the program.

do a logistic regression and pick a point along the ROC curve as you see fit (depending on what your precision/recall needs to be )

This really isn't the right forum for this type of stuff (try stackexchange). You are actually getting at something very deep: how can we extract causation (i.e. "variation in advertising spending causes variation in number of new clients per month") from correlation? Try correlating # of new clients per month against advertising spend "X" months ago (i.e. lagged). If you want to be adventurous and take into account the effect of other factors (to lessen Simpson's paradox http://en.wikipedia.org/wiki/Simpson%27s_paradox), try a multivariate Poisson regression (again with lagged factors).