Minimum Pubs for a PhD in CS? | Hacker News Reader

Minimum Pubs for a PhD in CS?

blog.regehr.org

37 pointsjnazario36 comments

My favorite is the rule of 7:

You're ready to graduate when the number of first-author publications you've published, plus the number of years you've been in grad school, equals 7.

Tongue-in-cheek, obviously, but not a bad guideline.

It's hard to say. I published 8 papers before submitting mine - 6 conference and 2 journal. I guess technically you don't really need to publish any, but doing so yields a few key benefits:

a) A conference paper early in the process can get you some really great feedback on your research. You might get to network with some people who are influential or prominent in your field. You may also get greater clarity for the direction you want to take the research.

b) Publishing a few papers means that a lot of the work you'll be submitting in your thesis has already been scrutinised by experts in much the same way that your thesis will be graded. If there are any weaknesses in your work or literature you've overlooked then this should come out in the review process when you submit your papers so you then have time to address the issues. Once published, you have greater certainty that the quality of your work is PhD suitable.

c) You get to put a timestamp on your work in case someone does something extremely similar in the meantime. It's improbable, but it's not impossible. If your ideas are great, the likelihood of this happening increases. If you publish stuff, you can point at your published work and say "look - this idea was mine". This almost sort of happened to me. I took it as reassurance that what I was doing had some merit.

d) Paper writing is similar to thesis writing, albeit on a much smaller scale. You still have to review the literature, present your work, evaluate it properly and do all the necessary comparisons to other existing stuff. If you can get a small paper published it's great practice for what you'll ultimately have to do for the thesis.

I can tell you from experience that usually there's not any written/formal publication requirement but I've never seen anyone getting his PhD in CS without at least 1 international publication (conference paper not workshop).

I disagree with people commenting that you could earn your PhD with zero publication. You need to compare your work with the state of the art and confront other people in conferences this is what research is about and not simply working on something and never open your work to critics.

In other theoretical field it's different but in CS there are plenty of opportunity to publish your work. PhD symposiums, poster sessions, workshops and millions of conferences (with probably only 10/20% of very good ones). So even if there's no formal requirement, I'd really suggest you put some effort into it. It's not just for the sake of showing who has the biggest (publication list), it's really to make you interact with people and have an objective opinion on your work through the review process of your submission to conferences.

I personally got my PhD with 11 publications (4 top tier conferences) and 5 patents. I can tell you it was a big shock when I submitted my 1st paper and got demolished but I really got valuable feedback I was on the wrong track and my advisor didn't care ;)

3 full papers at top-tier conferences in your field (you are the first author and the papers didn't have a bazillion authors). That's the unspoken standard I've seen in my area (systems). If you have an OSDI/SOSP paper, you could get away with less. I.e. One best paper at SOSP probably gets you a PhD as long as you were the primary person on the paper other than the advisor.

If you are looking for a rationale why 3 is a good number, then I'd have to say:

1) It shows you are not a one trick pony. This is essential if you are applying for faculty positions.

2) It is the number employers expect to see. This is really determined by the number of papers had by your competition. In this regard, the reason the number is 3 is because other people have around 3 papers.

This said, no one stamps a PhD on you when you get 3 papers. I actually got 3 papers pretty early in my PhD. I spent an extra year working with a research lab on an unrelated project. I found my PhD committee (full of some famous people) was not impressed by my papers. However, when I had some decent job offers, I felt I grew up in their eyes. So it is really about when you grow up, which doesn't happen by magic when you cross a certain number :-)

How did the fact that you have job offers come out to your committee? Is that part of the normal process to ask?

I imagine my advisor or another faculty I worked close with may have mentioned it. This sort of news gets around surprisingly fast in academic circles.

> Minimum Pubs for a PhD in CS?

The answer is obviously "one". Any number of PhDs have been granted based on one outstanding paper. But this outcome obviously depends on the quality of the work, and institutions differ in what they consider significant.

No self-respecting person/institution is going to say "N papers are required". The nature of the work is what's important, not a paper count.

It is not that obvious. Especially in theory, it is quite possible that you might have a thesis but not have published work anywhere or have things in the process of being published at the time of defense. I am not sure whether this is a good idea from a career point of view though...

A PhD is virtually never granted based on unpublished work. It might be tentatively offered pending publication, but not granted.

That depends entirely on your field. A (not insignificant) number of PhD students who've gone before me at my university got their PhDs without having published by the time of their viva.

Exactly. I believe the issue is that the culture in C.S. is to focus more on conferences which means a more rapid turnaround time. There are other disciplines where conferences are not as respected as journal publications which are slow as hell. Having said that, personally I tend to believe in the existence of life after a PhD, this means that one should probably publish and also give talks in conferences etc.

Not to go off-topic, but there are absolutely programs (and some may even be CS programs) which require "3 major papers" in lieu of a doctoral dissertation for award of the PhD. This surely has its disadvantages, but I have known colleagues--especially if they're headed into industry--who feel this is more representative of the type of work they're expected to perform and gives them a leg up on students who must manage both.

Be that as it may, I think counting papers is an absurd academic standard. It replaces evaluation with enumeration.

At my school (UCLA), there was not a formal requirement for publications or even an informal requirements. It looked like some professors emphasized publications more than others. I think, even if you aren't going into academia, they make a good unit of work and source of external feedback/validation.

My school did replace their Qualifying Examination with a research paper requirement (http://www.cs.ucla.edu/academics/graduate-program/graduate-s...), but it does not have to be published.

Unless things have changed recently, you certainly need more than 3 good publications to get a tenure-track academic position in CS. That means you have to either be a publishing machine while you are a student or do a post-doc.

A necessary distinction is between a "normal" thesis and a "thesis by papers". In the latter, the thesis is really just and intro and conclusion wrapped around already-published papers. In some universities in Ireland and in some other European countries, the student can choose either. In the case of thesis by papers, the required number of first-author journal papers is 3.

I haven't heard of any special dispensation for the case where the student has 1 paper proving P != NP. But the student is not obliged to choose "thesis by papers" so there is no problem.

The minimum number should be 0.

Papers are useless to rate the quality of a PhD thesis - the jury, if chosen right, should be experts in this field and thus able to analyze by themselves whether the PhD was an original contribution to science ie finding a way of doing something that was never done before and believed to be impossible, or in reverse, proving something to be impossible, and therefore PhD worthy material.

The amount of papers doesn't matter - or should not matter, if the PhD is to have any sense. Meaningless papers are just a waste of trees.

For ex, in my PhD I identified a specific kind of variability in written sign language that for some reason hadn't be grasped by the linguists who worked on the same problem with us. The PhD was in computer science and there was a linguist in the jury- she had read the acts of the conference where I had presented a communication on this variability, and for her this contribution alone was worthy of a PhD.

In the end, since there were other interesting things, it was awarded summa cum laude, even if not a single paper was published on the subject in a journal - my work was only disseminated in presentations, acts of conference, and a research report ordered by the government.

That was in 2008. 4 years later, my work is leading to an actual implementation of the ideas discussed, for the reasons discussed in the manuscript, and I have my own page on the SignWriting website (SignWriting is the written form of sign language I worked on) : http://www.signwriting.org/library/history/contributors/Azna...

I feel damn proud to have been proven right not just by some research paper published in an obscure magazine, but by real people working on this same problem and bringing the issues in front of the Unicode committee (in progress - hopefully it will be completed in 2013 or 2014)

It's far from done yet - it is still research work. But after my work from 2008, I made a tiny dent in the universe (as in http://www.openculture.com/2012/09/the_illustrated_guide_to_...)

When it's complete, I'll be happy to write a paper presenting the difference between the theory and the actual implementation, which could prove useful to other people attacking similar problems.

But a paper is not an end by itself.

EDIT - adding some links

Intermediary work : http://www.irit.fr/-Publications-?code=2628&nom=Aznar%20...

Manuscript : http://www.irit.fr/~Patrice.Dalle/Publis/these_aznar.pdf

I wholeheartedly agree with you.

Sadly, the state of academia has become a giant publication fest, where getting as many publications possible out the door every year has pretty much become a synonym with funding, tenure, etc. (if I check my spam folder these days, I have more spam about cheap conferences where you can get your paper published for $XXX than ones for viagra)

Professors pushing their PhD students to publish often are preparing them well for a future career in academia, if they so choose (which a significant portion of them will).

It's definitely not right, but it's the sad state we're at; and I'm not sure what exactly will change it :(

> are preparing them well for a future career in academia, if they so choose

Assuming that option remains. In other disciplines, from what I read, it is fast disappearing for the majority of... "qualified" candidates.

As the quality of the product (i.e. potential employee and what they have to offer) declines, I speculate so will academic opportunities. There's a lag while the decline in value reveals itself, but eventually the world catches up with the new state of things.

The minimum number should be 0.

In most places, it is. That said, I think most CS PhDs -- and definitely most strong CD PhDs -- will have a few publications in top conferences. This has several benefits:

(1) communicating your results -- few people read entire theses.

(2) getting feedback from the community, both in the form of PC reviews and feedback from the conference attendees.

(3) as a way to checkpoint your work on individual sub-projects, and to act as a forcing function.

Trying to "require" independent publications as part of a PhD seems silly, but having publications is definitely a good thing IMHO.

I fully agree with the benefits of communicating one's result, if only for feedback and getting independent verification or validation - but a paper is only a given way to reach this goal.

If your area of research is obscure or quickly evolving, a journal might not be the best place to share your results, while giving a presentation at relevant conferences, where people working on similar topics will get exposed to your result, is better.

But even this is secondary. I loved doing research, but I was even more interested in the end result - having something working and useful.

Not publishing might have damaged my "profile" in several places where the amount of dead trees is the only thing that matters, but it's all a matter of priorities - and I'm not sure I'd have been happy to work in a place where the end result is only an afterthought.

Maybe that's why I'm reading HN :-)

I think it's worth noting that in the more systems side of CS, researchers focus on publishing in conferences. Conferences have their own sets of problems, but the turn-around time is much better than that of journals.

Peer-reviewed publications (journals or conferences) are verifiable proof of quality. The higher impact the venues of publication are, the more qualified the papers.

I think a main tension here is between the idea that PhDs are awarded to a person versus to a dissertation.

On one extreme, if a student has published (say) dozens of top-tier papers and (say) demonstrated broad expertise in the field, haven't they earned a PhD?

On the other extreme, if a dissertation makes a huge advance in its field, doesn't that work deserve a PhD regardless of the author's other accomplishments?

Not always.

For example, if the dozens of papers are all disjointed, or on completely different topics, you can simply bolt them all together and call it a PhD thesis.

Like all good stories, there has to be a solid start, middle and end. This generally also means that one fantastic paper is also not enough, content wise, to warrant the award of a PhD.

What does it mean by a "good paper" (in CS) ?

- published at a high impact journal/conference ?

: There are gazzillions of paper out there in many good journals/conference proceedings but no one really cares about.

- total number of citations ?

: This may be a factor. But I have seen many survey papers (most of them are boring and mostly involve physical work) get a load of citations, but do not always add anything new.

- On the other hand, I have seen a number of good PhD thesis (especially in CS theory/math) that did not produce any "publication", per se. But they are highly cited and pretty much popular. Even many of them are written in Russian/German.

The impact of a "research work" is not always possible to be measured by some numbers. Getting the attention of the community is a totally different aspect -- and that's what makes a research work valuable.

No university in the UK has a "paper requirement" that I'm aware of. You're encouraged to publish, but it's not mandatory. Published work does increase the odds of successfully completing the process, because it pre-vets the work, but I've known people to finish with only one paper (or even zero). I'm writing up my thesis now, and have only 1 paper (a workshop, so really only 1/2). I hope that most of my work will be published post-thesis.

Of course, UK PhDs are very different to the US - here, 3 years is a normal timescale to completion, and 4 years is the maximum my department allows. Frankly, in any "deep" subject, that's barely enough time to work out what the state of the art is, let alone advance it.

I agree with this but also found that there can be a significant difference depending on your subject matter. It's typically easier for people in CS to publish because there are fewer barriers to developing and testing your work. Compare this to something like biology where a PhD student may have to cultivate an experiment involving animal tissues that may or may not become compromised somehow (for example), or some other delay that's much less likely to happen when working with computers, and you can see why they might take longer to develop something publishable.

Consequently there is an expectation that CS PhD students should publish more.

There is no minimum requirement in most universities in US. Most good PhD advisors however do have a minimum pubs requirement. Typically, in non-theory areas, it is 3 first-author top-conference papers, and higher if you are directly applying for faculty positions. The reason is that if you meet this requirement, they feel comfortable letting you defend yourself against the committee. (A student failing a PhD defense would be a big embarrassment for the advisor).

That said, such requirements do not matter to superstar candidates, their bar is much much higher. Typically, by this time, they are already giving invited talks at famous venues, so defense is merely a formality for them.

For many CS undergrad programs you don't HAVE to know any programming languages to graduate. You would just be at a serious disadvantage for many likely future positions if you don't.

Maybe you can finish your PhD without publishing but it seems unlikely that you could spend that much time doing GOOD original research and not have your name on something. Also it will be very hard to get an academic or even industry research position without GOOD publications.

I have to agree 3 is considered a good rule of thumb but it is all very dependent on a number of factors such as your individual department and advisor.

It varies by sub-field within CS and depends on the quality of the conferences or journals the papers were published to.

3 was the rule of thumb 20 years ago - but it was a guideline rather than a hard requirement.

> I'll omit my own views on this and maybe write about them later on

why not put your own opinion, Dr. "Associate Professor of Computer Science" ?

Because he doesn't want to poison the well. You get more impartial answers by asking without putting in your own opinion. "Hey, what do you think about X? Personally I think it's really dumb, but what about you?"

> Minimum Pubs for a PhD in CS?

This certainly exceeds acronym/real word ratio in my book. I thought this was about a university pub crawl... :S

Reply on news.ycombinator.com