Wednesday, March 29, 2006

Choosing Graduate Schools

An anonymous commenter asked
Many of us seniors are currently choosing among PhD programs. As you probably know, we are expected to come to a decision by April 15.

Would you mind sharing with us your advice on how an aspiring theorist should go about making this very difficult and important decision? I personally would find such a post very informative, and I'm certain many others would agree.

A great question but one where I have a conflict of interest—In my view you should all have Chicago as your first choice.

So instead I will post this advice from Bruce Maggs (from a 2001 interview via Higher Cohomology).

Choosing an university can be sometimes easy, sometimes hard. You may choose a university because there's a particular faculty member that you want to work with. That's risky, because often there's no guarantee that the faculty member will be able to take you on as a supervisor. As a general rule, it would be best to choose an university with a reputation for high-quality research results. This can be measured, well, by your opinion of different research papers. If you look at some papers, and you find some that you think are good, look where those authors are from. This may be difficult for a student who has yet to begin a research career: then the advice of faculty members at your undergraduate institution can help. I think it is generally a good idea, although I didn't follow this advice, to study for a graduate degree at a different school than your undergraduate institution, because you'll get a different point of view from the faculty, as you will meet many new potential collaborators. In choosing a graduate school the most important thing is that you understand what it is that you want to study.
Let me add that you should, if all possible, visit the schools you are interested in, talk to the faculty and current students. Make sure that you will feel comfortable in that environment, it will make for a much more enjoyable graduate experience.

57 comments:

  1. I think the best advice is somewhere in between what Bruce says: pick a school where there are two or more advisors you can imagine working with. Hope for your first choice, but don't bet your graduate career on it (since anything can happen with your first choice).

    I suppose the exception would be if you have contacted your first choice advisor beforehand and gotten some commitment that he/she will take you on as a student. Even then, something can always happen but at least the odds are in your favor at that point.

    ReplyDelete
  2. personal bias that no one cares about: in real sciences, the schools pay for your visit. in fact they have real interviews that actually make a difference. in comp sci, you will only know if you are admitted, and then you usually have to pay your own way to visit.

    just another barrier for poor/disadvantaged students in comp sci. another reason you see so many 'where have the students gone' blog posts by comp sci profs.

    i have a thousand other complaints but no one else cares about those either.

    ReplyDelete
  3. I second the vote for Chicago.

    ReplyDelete
  4. All the CS grad schools that I visited paid for me to visit, and I'm pretty sure this is the norm.

    ReplyDelete
  5. How important is the reputation of the school, outside your particular area? If you get into school XYZ that's universally very good and good in your area, or school ABC that's not as universally good but better in your area, which should you choose? In the short term, ABC seems the better choice. But I feel that, when it comes time to find a job, this may no longer be the case. In all the randomness of (academic) hiring, it seems school reputation can make a huge difference: for current faculty outside your area, how do they know ABC is especially good?

    ReplyDelete
  6. One thing that worked well for me when going through the process - find professors that had attended the schools I was considering and ask them about their experiences. Yes, this information can be out of date, but it gave some idea of what questions to ask currently-there people on visit days and after. Of course, this only worked for some of the schools I was looking at...

    Also, when looking I tried to get a feel for how quickly graduate students engage in research. At some schools, I found that there was a heavy emphasis on passing a set of exams during the first year or more of study. Until you passed these exams, professors were reluctant to work with you on research, as they didn't want to distract you from the exam...and the exam had a high failure rate. That did not make me feel more positive about those schools.

    ReplyDelete
  7. Just how important is it to graduate from MIT/Berkeley to get a good job in theory?

    I have the opportunity to go to Berkeley, but felt more comfortable at CMU during the visit days. If I want to do theory, would it be a mistake to go to CMU, which I gather is less well regarded in theory than in the more applied areas?

    ReplyDelete
  8. Having been a victim of adisor-go-round due to budget cuts at my university I would also suggest checking out the school's library.

    Much of graduate school is reading, and having a strong library with a good inter-library loan program is a must.

    ReplyDelete
  9. Anonymous I asks:
    How important is the reputation of the school, outside your particular area?
    Being for the first time on the other side of things - I am a prof at a top-20 (but not top-10) school, graduating my first PhD student and evaluating faculty applicants - I can tell you that when the market is down, the reputation of your school is (unfairly) extremely important. Even when the market is good, graduating from a name-brand school gives you a leg up.

    Anonymous II asks:
    Just how important is it to graduate from MIT/Berkeley to get a good job in theory?
    If you are comparing MIT/Berkeley to a top-20 school, see above. But you were comparing to CMU, where there are some very good theory profs (and it is a top ranked dept.). Still, the quality of the grad students is an issue, and the theory grads (overall) are better at MIT/Berkeley. But it also comes down to whose research you find more interesting...

    Caveat: the above assumes you are ultimately looking for an academic job. I think name recognition of the school is somewhat less critical if you are interested in industry/other career paths.

    ReplyDelete
  10. How important is the reputation of the school, outside your particular area?

    Reputation is, as someone else said, extremely important, sometimes unfairly. It is difficult to generalize as each department hiring practices are different, but a mediocre top-four graduate will generally have less resistance to land an interview in the top 20 than a star from a school ranked 15-20, even though the latter is statistically likelier to outdo the former.

    As general advice go to the best school that admitted you. However be mindful that there are situations and reasons in which this advice doesn't apply, such as funding, particular interest on a specific area that is not offered everywhere, visa issues, family reasons, etc.

    ReplyDelete
  11. Are the top four equal in this regard?

    ReplyDelete
  12. Are the top four equal in this regard?

    In terms of reputation I'd say generally yes, although there is some variation that is area specific. Plus, once you narrow it to the top four there are other considerations that come into play. Berkeley seems to do a better job at getting people academic jobs, while MIT seems to do better in getting your research going and having students publish rather early.

    ReplyDelete
  13. I think in theory the ranking is as follows:
    1- MIT
    2- Berkeley
    3-CMU
    4-Cornel
    5-Princeton
    6-Stanford
    What do you think?

    ReplyDelete
  14. I'd place Princeton and Cornel tied at number four and I think Stanford should be even lower. A quick check at their faculty page shows only two theoreticians: Leo Guibas and Tim Roughgarden. They are both top notch, but they are only two.

    ReplyDelete
  15. I think that a more realistic ranking is probably

    1. Berkeley/MIT
    2. CMU/Cornell/Princeton/Stanford/UW

    where the ranking within each group depends on the kind of research you want to do, and the kind of location in which you want to live.

    One shouldn't forget nearby resources like IAS for Princeton, Microsoft Research for UW, and IBM/Google/Yahoo/MSR for Berkeley and Stanford.

    If you get into MIT or Berkeley, it really is difficult to justify going somewhere else, especially if your goal is to get an academic job later. On the other hand, there are very good reasons why a person might choose any school in group #2 over any other.

    If you want do complexity, for instance, then Princeton and UW seem like better choices (with Princeton being a favorite in this regard). I think that where you get to live is an underappreciated factor, e.g. Seattle is probably nicer than Ithaca.

    ReplyDelete
  16. Stanford lists everyone in the world on their theory page, which makes it difficult to "count" the number of theory faculty. If you consider just "STOC/FOCS theory" then you have Goel, Koltun, Roughgarden, and Saberi (though 50% of these people are actually in "management science"), with Boneh and Motwani being very theory-friendly.

    ReplyDelete
  17. These four people in Stanford are junior faculty which are not comparable with thoese in MIT/Berkeley/CMU/Cornell.
    Also UW is not really among even top 10. GaTech, Yale, Harvard, etc are better.

    ReplyDelete
  18. "size" and "quality" of theory groups by highly objectionable measures.

    format:

    school #-theorists (avg. quality)

    #-theorists = # of people in
    theory group with focs/stoc paper
    in last 10 years

    avg. quality = avg no. of focs/stoc papers per theorist per year over last 5 years.

    Berkeley 6 (1.06)
    GTech 8 (0.825)
    Harvard 3 (0.8)
    Princeton 6 (1.23)
    Stanford 5 (0.92)
    UWash 4 (1.2)
    Yale 5 (0.48)

    *yale includes spielman
    *stanford includes goel,saberi
    **both footnotes improve avg. quality

    ReplyDelete
  19. Wow, the above anonymous has a LOT of time on his hands

    ReplyDelete
  20. a budding complexity theorist6:22 AM, March 30, 2006

    While I was deciding where to apply, I got the definite impression that Chicago is highly underrated in theory, particularly with the presence of TTI-C.

    This goes doubly for someone primarily interested in complexity. I was several times as impressed with the faculty at Chicago than I was with those at, say, Stanford.

    (thus, in a year or two when I apply again, assuming my interests are at all similar then to what they are now, I'll almost certainly apply to Chicago, and I probably will not apply to Stanford. While I'm at it I won't make the half dozen mistakes I made this time around.)

    ReplyDelete
  21. ...makes it difficult to "count" the number of theory faculty.

    That alone should tell you a lot. Stanford is strong in databases not because they have DB friendly people or people in other faculties who also do databases. They have strong DB people right in the CS deparment. Theory@Stanford hasn't yet recovered from the double whammy of Knuth retiring and Motwani migrating to other fields. Knowing how strong they are in other areas, I fully expect them to patch this deficiency within the next five years or so. In the meantime theory students with offers from the other top 3 departments might be better of not going there.

    Also UW is not really among even top 10.

    UW as in University of Washington is and has been in the top ten for over a decade in most departament rankings.

    While perhaps Chicago might be underrated among undergrads elsewhere, I do not think this is the case within the TCS community. Their reputation is solid and people like Lance, Janos and Babai are well known and respected. They even have a journal: "Chicago Journal of Theoretical Computer Science". The TTI effort is quite young, but already very promising.

    The "number of papers" metric doesn't work at these rarified levels. Paper count works to separate all departments into 4 or 5 basic tiers, but within each tier you need much finer measures to distinguish quality.

    ReplyDelete
  22. Finding an advisor you can work well with seems to be by far the most important factor.

    So here's a question for everyone. If chosing between X and Y came down to the following, what would your choice be? A young, energetic, prolific (say 3 or 4 STOC/FOCS/SODA papers per year) but pre-tenure advisor, or a well-known, established (say serving on multiple PCs or perhaps even the PC chair) and post-tenure advisor?

    Just how strong is the correlation between an advisor's placement record and her seniority? My guess is: Extremely strong, perhaps stronger than it should be. Not at all unlike the correlation between a department's reputation and its placement record.

    Anyone out there who can speak from experience to provide evidence for or against the above hypothesis?

    ReplyDelete
  23. A young, energetic, prolific (say 3 or 4 STOC/FOCS/SODA papers per year) but pre-tenure advisor, or a well-known, established (say serving on multiple PCs or perhaps even the PC chair) and post-tenure advisor?

    All else being equal, I'd take the post-tenure advisor. They bring higher standing and a wealth of knowledge---academic and otherwise---that they can pass on as advice to you.

    However this is not always necessarily the case. The old guard person might be working on area that is waning in importance, or s/he could be too busy doing PC work and not dedicate any time to you. It is also much likelier that in practice you will be supervised, at least at the beginning, by one of their senior grads students or PDFs.

    ReplyDelete
  24. I noticed that not a single person has mentioned Toronto so far. I guess nobody has heard of Cook, Borodin, Rackoff, Fich, Molloy and Pitassi, not to mention Corneil, Hadzilacos, Toueg, Libkin and Magen.

    ReplyDelete
  25. Cook, Borodin, Rackoff, Fich, Molloy and Pitassi, Corneil, Hadzilacos, Toueg, Libkin and Magen.

    Well, between the 11 of them they only had two papers in the last STOC/FOCS so they can't possible be any good... =:-]

    Seriously, I don't think there was an intention of being exhaustive in the listings. Many good theory places are missing: Duke, Toronto, Chicago, Rutgers, Harvard, SUNY Stony Brook, Columbia/Courant, Saarbrucken/MPI, Waterloo, UCSD and an untold number of universities and institutes in Israel.

    ReplyDelete
  26. I think it all comes down whether you want to optimize your graduate school experience, or optimize your chances for a faculty position after graduation.

    The fact is, graduating from a highly regarded "prestige" institution like MIT or UC Berkeley will open a lot of doors, and, all things being equal, it will help you more than going to UChicago or UCSD, even if you work with Lance or Russell.

    People unfamiliar with your area will be familiar with the prestige of your institution, and that strongly colors their first impression, and some places want to convert that to their own prestige (ever notice how some schools put the PhD granting institution under each name on their faculty list?).

    So, if there are enough people you could imagine yourself working with, go to the better ranked school.

    ReplyDelete
  27. If I was going to graduate school now and interested in Theory, my favorite would be Princeton. They're doing great, great research. They're the best! My second favorite would be Berkeley. They have a few more people, many more smart graduate students creating a fun environment for research, they have great taste in choosing research topics, and, well, let's not forget the social life... MIT would be a distant third since, from the outside, it seems too competitive for my taste.

    The other places: my choice would all depend on my particular interests and on how friendly the particular faculty matching my interests are.

    PS- I am not at Princeton.

    ReplyDelete
  28. Two other schools that people have overlooked are University of Maryland (Gasarch, J. Katz, S. Khuler, D. Mount, Aravind Srinivasan) and UCLA (Meyerson (asst. prof), Ostrovsky, Sahai).

    ReplyDelete
  29. Many good theory places are missing: Duke, Toronto, Chicago, Rutgers, Harvard, SUNY Stony Brook, Columbia/Courant, Saarbrucken/MPI, Waterloo, UCSD

    Umm...how about the University of Maryland? (Yes, I do go to Maryland.)

    ReplyDelete
  30. I was not at Princeton either, but I second the last anonymous. If you're interested in mainstream STOC/FOCS theory, Princeton is the place to be, as things stand now. Arora, Charikar, Barak are all extremely productive, and there are many other excellent theorists. Plus you get to work with folks at IAS.

    Cornell and Stanford don't even satisfy the basic requirement for a top 5 ranking - neither has an active complexity theorist...

    ReplyDelete
  31. Of course, rankings are area-dependent. If the applicant has a rough idea of which area he wants to work in, and is applying to schools in the U.S. and Canada, the following is a rough guide:

    Algorithms - Princeton, Cornell, MIT
    Complexity - Berkeley, MIT, Princeton
    Learning - CMU
    Cryptography - MIT
    Logic - Toronto

    ReplyDelete
  32. Cornell and Stanford don't even satisfy the basic requirement for a top 5 ranking - neither has an active complexity theorist...

    Are you talking about theory or "complexity theory"? A school doesn't need a complexity theorist to be in the top 5 theory departments any more than it doesn't need a computational geometer to rank that highly. Of course good schools would tend to have both the complexity theorist and the computational geometer, but this is not strictly necessary.

    ReplyDelete
  33. Did Charikar et al disprove the unique games conjecture? There is a paper called "How to play any unique game" on his students' website but no actual paper ...

    ReplyDelete
  34. No, they did not disprove the UGC.

    ReplyDelete
  35. Anon 3:57 - Cornell has one of the best algorithms groups in the world, which makes the lack of a complexity theorist stick out even further. And to think that this is Hartmanis' school...

    Comparing complexity theory to computational gemometry is absurd. Computational geometry is an important area in its own right, but if we're talking about theoretical computer science as a whole, complexity and algorithms are at the center. Algorithms is always going to be a more active area, that's fair enough. But let's not forget that the most fundamental problems in theory concern complexity. No department that is indifferent to foundational issues can claim to be a top theory department.

    ReplyDelete
  36. sadly, researchers are judged by number of stoc/focs papers. if this is all it matters, why waste time here? focs deadline is approaching...

    ReplyDelete
  37. It seems that ranking always causes drastic arguments. Is this undecidable or decidable?

    ReplyDelete
  38. It seems to me that a public blog aimed at the community at large is not really the place much of this discussion, especially with people throwing out specific, numbers, orderings, names, and opinions thereof. Even the original question might not have that much relevance here. I imagine students would be better off seeking personal advice from professors in the area at their undergraduate universities with whom they have a good relationship.

    ReplyDelete
  39. What makes the stoc/focs metric of achievement so absurd is that the entries are given only the most cursory "beauty contest" reviews, and totally bogus papers are known to crop up from time to time, often without so much as a note of errata...

    ReplyDelete
  40. By the way, if anyone is looking for grad schools in cryptography, I've got a little list:
    http://www.cs.berkeley.edu/~dmolnar/gradschools.html

    No rankings, just schools and professors. I'm sure it's a little out of date, so I welcome corrections and additions.

    ReplyDelete
  41. Anon 3:57 - Cornell has one of the best algorithms groups in the world, which makes the lack of a complexity theorist stick out even further.

    Cornell is no doubt a good place to do algorithms, although I think is a notch below MIT's algorithms group of Demaine, Indyk, Karger, Leighton, Leiserson, Shor, Sudan and Vempala.

    In terms of the need of a complexity theory person to be a top notch computer science department, currently MIT has no active full-time complexity theorist (granted, the complexity papers that their crypto group publishes on the side are as good as any's).

    To be clear, I'm not arguing that not having a complexity theorist is a desirable state. It is not.

    ReplyDelete
  42. How possible is it to transfer grad schools? My problem is that my current options are not as strong in theory as I'd like. Will getting grad course credits elsewhere strengthen an app?

    ReplyDelete
  43. How possible is it to transfer grad schools? My problem is that my current options are not as strong in theory as I'd like. Will getting grad course credits elsewhere strengthen an app?

    It is certainly possible to transfer. However, graduate course credits on their own don't do that much since grades in graduate courses are often taken less seriously than grades in undergraduate courses.

    What really matters is that you impress your recommenders, get involved in research, show promise of being able to do a Ph.D. and not just courses, and write a statement that shows you are knowledgeable about the field at a level beyond that of typical undergraduate applicants.

    ReplyDelete
  44. Speaking of rankings, the new USnews rankings just came out... The top 4 are of course the same, but Stanford has moved from 5th to 2nd in the theory rankings, surpassing Berkeley.

    Surprise?

    ReplyDelete
  45. Not really a surprise. With Goel, Saberi, Johari, Roughgarden, Koltun, Guibas, Plotkin and Motwani it's a strong department. Perhaps there're no complexity theorists, but there're two computational geometrists to counteract it :)

    ReplyDelete
  46. USNews theory rankings are pretty much worthless. Polling people in the field (undergrad faculty and their contacts) is a much better idea...

    ReplyDelete
  47. The funny thing is that neither the Hebrew university, Tel Aviv University or the Wiezman institute are mentioned, and all three are arguably way stronger than some (if not most) of the places mentioned on these lists. And all three have strong complexity people. (Yeh, I know that these places are in Israel, but if you really want to do good theory, you have to go to a good place.)

    In short, all ranking lists are stupid, but some lists are more stupid than others. Especially if they rank universities doing theory in CS and ignore whats going outside the US.

    ReplyDelete
  48. Probably you are from Stanford. Come on, we all know that we should not trust too much to USNews for these facts.

    ReplyDelete
  49. One should take the exact details of US News rankings with a grain of salt. While they are generally not terrible they sometimes have major problems, such as highly ranking certain graduate programs at name schools when those programs don't even exist.

    ReplyDelete
  50. The funny thing is that neither the Hebrew university, Tel Aviv University or the Wiezman institute are mentioned,

    They were mentioned already as in "an untold number of universities and institutes in Israel". And by the way you forgot Technion, which is also very strong.

    ReplyDelete
  51. Probably you are from Stanford. Come on, we all know that we should not trust too much to USNews for these facts.

    Not long ago the Oregon Institute was named #1 in DB by the USNews. This is a fine institution, but I doubt anyone outside the USNews would rank it number one.

    ReplyDelete
  52. You forgot Poland.

    ReplyDelete
  53. If you are going to get a PhD in Theory, you should go to the place where you will do best in graduate school. This is going to be an individual choice: programs are very different, and so are graduate students' modes of learning.

    Examples: how much coursework are you going to have to do? And what is good, little or lots?

    Depends on how confortable you are learning things on your own, and how well you work with only partial knowledge of an area.

    Do you go for prestige or excellence in your subarea?

    Answering with a question: how committed are you to the subarea? If you really know you love a particular subject, go to the place where you can do it best.

    Will I be at a disadvantage if I do not come from MIT/Berkeley?

    Yes, somewhat. But you will also be at a disadvantage if you are only number 2 in your cohort at Berkeley, and there is a clear superstar competing with you. People can directly compare you to her, as opposed to the situation where you are number one at a "lesser" place, like Toronto.

    Should I work with a less productive senior advisor, or with a junior faculty member who is incredibly active?

    Depends who you work better with. Work with the person who gets you to get better results, and who points you towards problems you really like, and you make some progress towards solving. If the junior person is really good, do not worry about his tenure: if the university makes a mistake, your advisor will probably get a job at another great university, and you'll finish your research there. Besides, your advisor might change institutions even if he is senior (Blum and Karp come immediately to mind).

    How important is the location?

    Depends. How much time will you spend outside books? How much do you want to go to the opera? Do you have a significant other?

    In short: grad school can be great fun. It is the time you really have very little else to do besides becoming a good scientist. Go all out to make the best of it!

    ReplyDelete
  54. As a related question: what's the best way to go about picking a thesis topic?

    ReplyDelete
  55. here is my suggestion:

    Algorithms - Princeton, Cornell, MIT

    Complexity - Berkeley, MIT, Chicago,
    Rutgers, Toronto, UW, Princeton

    Computational Geometry - MIT, UIUC, Harvard, Stanford

    AI - CMU, MIT, Stanford

    Cryptography - MIT, UCSD, Waterloo

    Combinatorics - MIT, Rutgers, Princeton

    Logic - Toronto, UCLA, Berkeley, UIUC

    ReplyDelete
  56. what will be the best universities in US to apply for a phd in computational geometry?

    ReplyDelete
  57. UK CS PhD programs? How do they stack up?

    ReplyDelete