Sunday, August 09, 2009

Publishing Goals - Diversity

For the first time, I'm a co-author on a graphics paper; the paper Real-Time Parallel Hashing on the GPU, with co-authors Dan Alcantara, Andrei Sharf, Fatemeh Abbasinejad, Nina Amenta, Shubhabrata Sengupta, and John D. Owens of University of California, Davis, was (conditionally) accepted to SIGGRAPH Asia 2009. More on the paper in a later post.

This post is about why I'm excited about having a graphics paper, when I've never had one before. Early in your career, when you are just starting to write papers, the goal for a typical student is to aim to get papers in the top conferences in your area: theorists aim for FOCS/STOC/SODA, networking people aim for SIGCOMM, systems people aim for SOSP/OSDI, and so on. It's the way to establish your professional reputation.

The downside of this is that as a student you become focused on both the style and problem spaces of your "home" community. Some (most?) researchers become entrenched in their home community, and never really leave it. My biased view, though, is that when I look at the great computer scientists and mathematicians that I think of as role models, they've had a broad impact across multiple areas, and have crossed boundaries to publish outside what might be considered their comfort zone.

For quite some time, I've tried to follow that lead. So one of the goals I have is to publish in a variety of different conference areas -- really, as many as I can. (I realize it's taking the logic the wrong way -- great scientists didn't necessarily set out to do things in many fields, but their work was so great it naturally had impact in lots of places.) It's an apparently strange goal, in that I haven't heard others voice it, but it's one I enjoy. I optimistically like to think that I can make contributions in multiple areas, and it gives me the chance to get some insight into things going on in different areas. It also leads to fun collaborations.

Of course, just publishing a paper in a conference does not make you part of that conference community. While I've found it rewarding to write a paper for a new conference I've never been to, I've also found that when I have a stronger interest, I can get more out of a longer relationship. My personal rule of thumb is that once I've had three papers at a conference, I can consider myself part of that conference community; at that point, generally I'll know people, and they'll know me, and there's a feeling that my work is really part of a larger scientific picture going forward. (I've also found that's about the point where people start asking you to serve on the program committee...) So my goal is not just to publish single papers in as many conferences as I can, but also to explore and find interesting communities to work with.

Overall, I feel this path has served me well. I feel like I have several research home communities (in theory, networking, and information theory), and when I visit someplace to give a talk or interview a candidate, there is often an unusual connection to be found. I think I've created a positive feedback loop -- because I'm known to have a wide array of interests, people feel free to approach me with new problems, which I'm happy to encourage! [This graphics paper arose entirely because Nina Amenta wrote to ask me about what's known about doing "modern" hashing on "modern" parallel architectures such as GPUs...] But there's certainly potential downsides. I know that, as I was coming up for tenure, this "lack of focus" (if one wants to label it negatively) was, at least for some, a problem to be overcome rather than a positive virtue. Graduate students and junior faculty may need to keep in mind how such forays will be viewed -- usually it seems to me a very small number is seen as a positive, but a large number isn't necessarily seen as a good thing.

I understand for most people the goal is to publish FOCS/STOC papers, or to publish SIGCOMM papers, or publish SOSP/OSDI papers -- that's what it means for them to do great work, and that works for them. My approach is different, so I'm excited by the novelty of having a SIGGRAPH Asia paper. While I wouldn't prescribe how people should do research, I'd encourage people to look for opportunities to work a bit outside their home community. And I think there's value in that that tenure and hiring committees should recognize and acknowledge.

And by the way, I've never had a SOSP or OSDI paper -- if anyone has any good ideas of how that might be rectified, feel free to let me know.


Avik Chaudhuri said...

Interesting approach; I kind of want to do the same thing in the long run. The first goal, of course, is to bridge theory and practice, even within the same field.

Anonymous said...

I like your approach. Unfortunately, my experience is that you end up being called a dilettante which i think is very sad. Especially given that the fields are so narrow and specialized that interesting ideas from neighboring fields might not only come in handy but are necessary for further progress.

David Andersen said...

Hi, Anonymous - if it's the kind of research you want to do, who cares what people call you? Yes, you'll face a challenge of spinning it appropriately for the tenure committees, but life's too short to do other than the work you love. And if doing interesting work that applies to a variety of fields marks one a dilettante, then sign me up!

Btw, the SOSP paper list is also online now, with links to most of the papers. I'm not sure there's such an obvious keyword as there is for SIGCOMM -- I suspect that the identifier for this year's SOSP papers was "tons of time put into the papers and the implementations behind them". Tom and John collected some great stats about the # of person-years that went into the average SOSP paper this year, and it was surprisingly high.

Anonymous said...

You said that about three papers is the point where people start asking you to serve on the program committee? Perhaps it is so in other areas, or if you are already well-known. I think that in TCS things are different.

Michael Mitzenmacher said...

Anon #4: A smaller number of papers in other communities before being asked to serve on a PC is both a function of being already known in another area (I'm sure that played a part in my case), but also just differs among communities. Keep in mind SIGCOMM takes about 30 papers a year, for a community much larger than TCS; by the time you've got 3 SIGCOMM papers, I think it's generally (and rightly) assumed you're ready to serve on networking PCs.