As far as I can see, the top PhD programs are getting better and better applications for the same number of slots, and the process is becoming far more competitive. There are plenty of students coming in with publications at good venues (maybe not NIPS, SOSP, STOC level). It used to be that students came in without doing research, but that doesn't seem to be the norm any more.
I don't know how I got into grad school --- looking at the qualifications the incoming students have these days I don't think I'd make it --- and I only started my PhD 4 years ago.
This ratcheting up in qualifications is happening at every level in academia (and not just in CS). I know some grad students whose work and lifestyle is best described as "tenure track professors in the 80s." I know researchers at top labs like MSR who were hired on the strength of a _single_ OSDI publication. I had one as a second year PhD student, and I'm so far off the ball that I don't think I will even be attempting to go on the academic job market when I graduate. At this point I think it's pretty fruitless to apply to do a PhD without having prior research experience, ideally with a publication. I may be getting too jaded and pessimistic though. Consider alternate points of view.
On the other hand, if you can get in, it's a pretty amazing life of learning and developing industry relevant skills (if you're smart about it) on the NSF's dime. I know some of y'all are cynical about academia and like to badmouth PhD students as being unhirable and writing crap code, but consider that RTM is a top academic, and is pretty relevant as far as the startup scene is considered.
Matt Welsh, Eric Brewer, Jeff Dean, Sanjay Ghemawat, Mike Dahlin all have PhDs and lead teams that write serious production code at Google. Ion Stoica and Hui Zhang have started technically heavy networking startups. Joe Hellerstein has a new startup (I'm not familiar with the details).
You don't hear much about them because they aren't sexy consumer facing startups of the type that HN is obsessed about. They typically focus on unglamorous but technically challenging problems of the kind that require deep expertise.
Add Tom Leighton (Akamai), Frans Kaashoek (Mazu Networks), Dawson Engler (Coverity), and maybe David Brumley (ForAllSecure) to that list. Profitable startups are not uncommon among professors in top departments.
All the other companies I listed have had either successful exits, large fundraising rounds, or a successful IPO. David's company is much newer and doesn't meet that bar yet.
Some more, fairly randomly chosen: Peter Norvig (USC->Berkeley->Google), Chris Lattner (Illinois->Apple, LLVM), Daniel J. Bernstein (Illinois, qmail/djbdns), Donald Knuth (Stanford, LaTeX), Hadley Wickham (Rice, ggplot2), the whole PLT Scheme/Racket team, etc.
I'm curious what you plan to do after finishing your PhD? Have you considered doing/working at a startup? I assume post-doc is off the table since you are not interested in the tenure track grind.
I'm not 100% sure. For now I'm concentrating on finishing a serious PhD (my plan is not to quit academia with a pity-PhD consisting of no serious publications). I'm working on a couple projects leading up to publication, and that doesn't leave much time for strategic planning :)
But for future plans I would very much like to work on some large scale projects building out large scale infrastructure. I don't have much experience writing production code for large scale distributed systems --- my research has, by necessity, concentrated on prototypes and theoretical innovations. I would be very excited to join a group that rolls out cutting edge systems, such as what Matt Welsh, or Jeff Dean, or Eric Brewer are doing at Google. I'm not sure a consumer focused startup would offer such an opportunity (although I'm willing to be convinced).
Here is some unsolicited advise from an old timer. As a backup, do one of your internships in a dev shop. It can be Google, Amazon, Facebook, whatever. The key thing is ... it has to be non-research. It is far easier to do this kind of exploration as a student than later.
"...a pity-PhD consisting of no serious publications" ... right... that would probably be the worst of all worlds, considering that the piece of paper doesn't offer any guarantees, so hanging around just to get the diploma would be risky.
All good PhDs are funded (if it isn't categorically don't go). Typically this is through advisor's grants, which are usually from the NSF. I'm an international student (originally from India) and I'm funded through my advisor's NSF grant (I think. I'm not 100% sure since I don't know the exact details and he also has some additional funding from other sources like DARPA that he could be using). You should take the fact that I don't really know where my funding comes from as a good sign that PhD funding is very stable.
It may be worth noting that in many areas of CS, you can still do research and think deeply about hard problems even if you don't go to grad school. You can read papers on the internet, which are free in some cases, and use your own computer. It will be a lot harder to publish anything, you won't have a faculty adviser to steer you in the right direction, and you won't get to go to conferences. But nothing stops you from doing research anyway, and if you happen to solve some important problem and can prove it, eventually word will get out and people will notice (one would hope).
A big problem is the lack of publication access though. I happen to have most accounts through work (ACM, IEEE, etc.), but if I didn't, I'd have to find other ways to get about half the papers I read.
I think the "unschooled" approach is more appealing for practitioners / systems software. BitTorrent and BitCoin both have had papers, though of course that's wasn't their real point, and not their real impact.
If you want to prove results, it seems rare to have the motivation outside the academic community. The environment and feedback is important.
But I'd be interested in any examples of people who have done this. I recall seeing some examples of professors without Ph.D.'s a few months back on HN.
I think anyone who is proving good theoretical results will be quickly offered an academic job, and take it, since they likely won't have another way of supporting themselves.
Yeah... I wasn't so much suggesting that it's a viable path to a career in research as observing that if you missed your chance to do a PhD, you're not cut off for life from the subjective rewards of research that Dr. Harchol-Balter alluded to, like the satisfaction that comes from working on fundamental problems. You'd need more self-motivation. However, when research isn't your real job, there's no pressure, so you can take your time, and produce results or not, as you wish.
If it's to advance the state of the art in academia, or to push an academic career, the Phd really is required. (Much of it is the habits and network too)
If it's art for art's sake, the Phd is not required.
If it's to advance a professional (non-academic) agenda, the Phd isn't required either.
In the latter two cases, it can help, but it's an expensive degree when you factor in opportunity cost.
If you find a topic interesting, then you will enjoy reading the papers on it. You will enjoy thinking it through, just to get a strong grasp of the research that's already been done. In the course of turning it over in your head, you may have an idea and want to follow up on it. It might work and actually be novel. That's research. It's probably not the kind that fundamentally alters the field, or could win best paper awards, but if it hasn't been done before and it sheds new light on something that matters in any amount, it's real. In some fields you can post things to newsgroups and web boards where bona fide experts will see it, in lieu of publishing in a journal or conference.
You don't have to set out to achieve something worthy of research. If you learn about something you're interested in, and you have an inquisitive mind, you might incidentally have an idea that hasn't been thought of. I think it helps to at least have a B.S. in the subject, preferably at a strong research school where you can see research up close and even do some as an undergrad. Undergrads can take PhD courses as long as the professor approves it, or at least that was true for me.
It was a very good article and I agreed with most of it except that people who decide to leave academia after PhD still have learned a lot: self management, self motivation and critical thinking as well as learning how to think are several of them. I would say that thinking of PhD as a wasted time assuming that the person is not pursuing academic career is something I don't agree with.
Pretty good advice, much of it not exclusive to CS alone. In particular the notes about the opportunity cost involved with getting a PhD.
I'd add to this: find out with absolute clarity upfront what the terms are for applying as a doctoral student vs. a masters student (with option to matriculate).
Often there is an expectation that doctoral students can and will pass qualifying exams within a pretty short period of time (a year perhaps) of being accepted to their program. This was the case in my own program.
I wish the opp cost was made more clear cut in some way. Maybe a rite of passage where you take a picture of a house and burn it or something. These days when an undergrad can pull 6 figures, a PhD is very, very costly.
I have BS in CS, MBA-FINANCE, neither will provide the real experience in a real hacker world, less will do a Ph.D a total waste of money and a debt that will hunt you for a long time.
Time=Money
https://www.coursera.org/
Sorry for the aside, this opened for me in pdf.js in firefox (on archlinux) and the font rendered very poorly, but when I downloaded it and looked at it in zathura (vi-like pdf viewer) the font looked fine. Does anybody know what I can do to fix the fonts in pdf.js?
The European system follows a strictly defined progression, LMD:
Licence: 3 years after high school, equivalent to a US BSc.
Master: +2 years on top of the previous, equivalent to a US MSc.
Doctorate: +3 years on top of the previous, equivalent to a US PhD.
Some PhDs can take a bit more time - it's not uncommon for a student to take an extra 6 months or so (or to start a few months after the official school year starts, because of funding issues). However, I've never heard of someone's PhD taking 4 or 5 years or more (although I'm sure it may happen in some edge cases).
The reason why Licence is 3 years compared to the US's 4 is that it's very focused: often, students don't pick which classes they take until the end of their curriculum (where you can pick classes for a specialty relevant to your discipline; for example, AI or Graphics for CS). There are a few general ed classes (for example macro economics, English as a second language, etc.), but definitely less than the US. Similarly, you don't get too pick. I've seen US students get a BSc. in Computer Science and having taken theater and anthropology classes in their first year; this does not happen in Europe (and you do notice that in the US, some very focused students do get their BSc. in 3 years).
This also means that the average euro student with a licence is typically stronger in her field than the average US student with a BSc.; however, the downside is that you can't change majors halfway or anything. If you do 1.5 years of CS and then realize you want to go do something else, it's pretty much back to square one.
The reason why a PhD is shorter is also due to the fact that it is common for the 2nd year of the Master (M2) to be split in two branches- the standard one, and M2R (for Master 2 Research) for students who know they want to go into research. M2Rs typically include a 6 months internship in a lab, where you start working on topics that will be related to your PhD (you can still go work after an M2R and not do a PhD, but it's uncommon).
Source: I've been through the French, British, and US university systems :)
In the UK, many of my peers finished somewhere after 3 years but before 5. The writing-up phase tended to drag on, I had a couple of friends who ended up working while writing up as their 3-year funding ran out.
Really? How so? I was a research assistant in a US lab, and we interacted regularly with European groups of PhD students/post docs/professors, there was never such a feeling. I have friends doing a PhD back home, and they publish at international conferences etc. just like their American peers.
Note that by PhD I mean the degree itself, not the person holding it -- I'm sure highly talented people do PhDs in Europe. The simple fact that a US PhD is more time consuming would be enough to account for a difference in esteem. Check out arjunnarayan's comments on this same topic on a parallel thread -- he has more to say about it than I do. I'm repeating what I've heard, and IIRC I heard it from CS professors where I went to college -- the same one where Mor Harchol-Balter works.
I don't mean to propagate bias. I don't really have my own opinion, I was repeating something I heard from insiders in the US PhD process. They no doubt have some home team bias. But how well the PhD helps you get jobs is something that's potentially measurable and testable.
European PhD students don't do as much research, or publish as much. They are not competitive with US PhD students for academic jobs in the US. This is not a function of their quality, but a function of how much time they've had. It's also more usual to do a postdoc in Europe for 1-2 years, at which point their publication list and resumé looks about the same as a 5-6 year graduating US PhD student.
Talking to european grad students, it appears they have lives during grad school. People talk about it like a job with fixed hours. The average systems PhD student at a top US or Canadian school lives in the lab.
As a former Berkeley grad student who managed to do ok (http://barnowl.org/research/) working normal hours (8/9 to 5/6, except around paper deadlines), I would dispute the claims here and in many similar threads that crazy hours are required for performing successful research (in computer science at least).
I would dispute the "getting more done" claim. One reason whay UK PhDs are shorter is that they typically don't have any classes or rotations. You dive straight into the research.
This is a best vs best test (as opposed to an average vs average test which requires more painful empirical effort), so focus on the top European PhD students. If you have methodological issues with a best vs best test because it doesn't not reflect the average PhD experience, let me know as I cannot think of any realistic reasons why it would be a bad test.
What's interesting is that you point these out as good examples, but in Europe these would typically be seen as evidence of supervisor malfeasance, or at least raise some eyebrows. If a student you're supervising has published this many articles in top journals and conferences in the field, has been in grad school for 4+ years, and has still not graduated, someone is going to wonder: why not? They clearly are both smart and accomplished enough to have graduated. So why haven't they? Are you milking them as low-paid labor instead of letting them graduate? Are you taking up all their time with grant projects unrelated to their dissertation, so they haven't had a chance to finish it? Etc. Questions get asked!
The idea is that the top grad student shouldn't be above a certain threshold, because if they are, they should have a Ph.D., not be artificially held back.
Interesting to note that in the UK it is now routine for there to be a four year deadline on PhD students. Where I work, instrumentation people were routinely finishing late because they were enjoying the lifestyle a bit too much. They're trying to streamline the system a lot.
We now have to jump through a lot of hoops - 9 month, 12 month, 18 month review panels. It's a pain, but it's obvious they're trying to make sure you don't spend a year in write-up mode. They absolutely want you through the system, because if you don't finish on time, the department gets a funding cut.
And actually you make a good point. In the UK, PhD students are fairly well paid. I'm on a CASE program and make around £17k a year tax free (outside London). The nominal wage is £13.5k for outside London (£15.5 inside). We're treated like research staff and not cheap labour.
That's kind of curious logic. If a student is so accomplished that it's plainly obvious to everybody that they deserve a PhD, then the degree itself is just a formality. What's stopping that student from leaving, if they want to?
I'm sure it makes sense in the context of European academia, which illuminates the contrast with the US.
I think it's slightly biased because the US has an extremely strong CS founding, it's world leading without a doubt. This is pushed by a large number of excellent departments and groups like Microsoft Research and IBM (hell, even Disney) which churn out top tier research.
You cannot judge people on long lists of conferences. The people you cited have enviable CVs but only three or four peer reviwed journal papers (prior to 2010) between them. This is comparable to anyone in the EU. What are the impact factors of those journals? I have no idea and am unqualified to judge. It depends entirely on your field and what the big places to publish are. For me, an imaging PhD student, it usually starts with ICCV and CVPR (and of course SIGGRAPH), however the actual citations come from journals, not the conference papers.
I have been chastised by my supervisor for citing proceedings before - a lot of people don't consider them peer reviewed and that makes a big difference.
I know people who got papers in Nature or Science during their PhD posts. In astrophysics, people regularly get four-five peer-reviwed papers over the course of a PhD because it's a publication-heavy area. A single Nature or Phys Rev citation is worth a lot more than five conferences you attended. I probably won't publish more than a couple of papers in my field, but I'm not expected to as some of it is commercially sensitive (though I may get a patent out of it instead - and more importantly, a job!).
At the end of the day, does it matter? Very few PhD students are genuinely top tier. Most of us are happy making a dent in our little niche and will probably get comfortable jobs afterwards. Unfortunately things inevitably become a slanging match :(
This is a big US/EU cultural difference in CS. In American CS, in many sub-areas, conferences are considered the top place to publish: CHI, SIGGRAPH, NIPS, etc. are basically the most prestigious place to publish, and are frequently cited. In some areas the journals are actually less prestigious: there is no HCI journal that matches ACM CHI's impact factor or name recognition. And many AI journals are considered worse then AAAI or IJCAI as publication venues (though JMLR and JAIR are well-respected journals).
But in Europe, it's more similar to the way it is in other fields in the US, such as physics: conferences are considered a fine place to present small bits of work, or work in progress, but not a proper archival publication for major results, which should go in journals. Journal articles get the main prestige and are what's expected on a CV, and citations to journals are strongly preferred over citations to proceedings.
This leads to an interesting situation where some prominent tenured professors in the U.S. would be weak candidates for European faculty positions, and vice versa.
Hence why I mentioned in my field, often the top publications are in ICCV/CPVR/SIGGRAPH and these are valued a lot more than many journals. As most of the work I review falls under computer science, I read a lot of highly cited conference papers. I'm also
I did get the impression that people tended to republish really 'big' results in places like IJCV or TPAMI. But, coming from a physics/UK background I guess I'm sufficiently biased towards journals!
In Europe you typically need a masters degree before you start a PhD program. In the US you don't - you can think of it as a 2+3 type program where you spend the first couple of years taking classes (and getting your masters) and the next three doing your thesis.
I always thought part of the reason was that students in the US basically spend 2 years in their undergraduate degree taking general courses. So US graduate students spend the first few years taking graduate courses rather jumping straight into research. There's lag. That's just my own speculation though.
I don't know how I got into grad school --- looking at the qualifications the incoming students have these days I don't think I'd make it --- and I only started my PhD 4 years ago.
This ratcheting up in qualifications is happening at every level in academia (and not just in CS). I know some grad students whose work and lifestyle is best described as "tenure track professors in the 80s." I know researchers at top labs like MSR who were hired on the strength of a _single_ OSDI publication. I had one as a second year PhD student, and I'm so far off the ball that I don't think I will even be attempting to go on the academic job market when I graduate. At this point I think it's pretty fruitless to apply to do a PhD without having prior research experience, ideally with a publication. I may be getting too jaded and pessimistic though. Consider alternate points of view.
On the other hand, if you can get in, it's a pretty amazing life of learning and developing industry relevant skills (if you're smart about it) on the NSF's dime. I know some of y'all are cynical about academia and like to badmouth PhD students as being unhirable and writing crap code, but consider that RTM is a top academic, and is pretty relevant as far as the startup scene is considered.