Sunday, July 06, 2008

ISIT this week

While many theory CS people will be spending the week at ICALP (and
adjacent workshops and such), the information theory people will be
spending the week at ISIT. No wonder these communities don't get
together as much as they should -- conferences are cross-scheduled!
CS theory will be well represented at ISIT, however, as Avi Wigderson
is one of the plenary speakers.

ISIT is bigger than, well, any theory CS conference I know, because it
is the major IT conference each year. CS theory has nothing
like it, with more, smaller conferences throughout the year, and (it
seems to me) many more specialized conferences and workshops. So here are some things I think worth observing for theory CS people, just to think about how we do things, and if we'd want to change:

1) ISIT goes for a week. One day of tutorials, 5 days of talks, with
4 parallel sessions, and a plenary each day. So naturally, most everyone
comes. (Um, no, I'm not going this year. That new baby thing...)

2) There's specific time for the Board of Governors of the IEEE
Information Society to meet and do their business. A benefit of a
conference where most everyone comes is that having "business meetings" like this seems easy to set up. Similarly, there's a large Awards Luncheon, and most people are there to pick up their awards, and see/hear the award-winners.

3) There are several activities especially for students. Roundtable
discussions, a panel led by the student committee, and a panel on
balancing career and personal life (see here and here for recent related posts on Sorelle's blog on that theme). Generally these are done over lunch, and lunch is provided for the students. (Never underestimate how well students respond to free food.)

4) Part of the tradeoff in establishing a large conference is that a higher
percentage of papers are accepted. And there's an understanding that individuals are not supposed to submit large numbers of papers, as large-scale participation is one of the goals. (This used to be explicit in the call -- something about multiple papers from an author being subject to more scrutiny -- but I don't see it in this year's call.)

As a relative outsider, I enjoy the ISIT setup. There's one
conference I know I can send my IT papers too; when I go, there's a
chance to see everyone, though there is sometimes the challenge of
tracking them down and scheduling a meet. There's a clear and strong
sense of community at the conference, despite the size.

I'm not trying to say that CS theory doesn't have a community-feeling.
But it does feel like the CS communities tend to partition themselves
more into loosely overlapping subcommunities. There's no universal
conference, although perhaps SODA (moreso than FOCS/STOC, based on sheer size) comes closest. I wonder, sometimes, what we as a community lose from this.

5 comments:

Anonymous said...

Also it turns out that ICALP is cross-located with COLT, the learning theory conference, and ICML, so the theory community and the learning theory community do not meet either.

David said...

I was just reading the SODA 2009 website. Is it typical in CS to have submission deadlines six months in advance of the conference? It strikes me that if you submit a final paper six months in advance, you [and the field] will be further along or further afield of it by the time that you actually present it.

Paul Beame said...

ISIT goes for a week. One day of tutorials, 5 days of talks, with 4 parallel sessions, and a plenary each day.

I am in the Toronto this week and stopped by yesterday afternoon. It was not 4 parallel sessions but 8 parallel sessions. This felt almost on the scale of FCRC, though the sessions and talks seemed less formal and there was a much wider range of talk quality both in content and delivery.

Anonymous said...

Michael,

You say "Part of the tradeoff in establishing a large conference is that a higher percentage of papers are accepted. And there's an understanding that individuals are not supposed to submit large numbers of papers, as large-scale participation is one of the goals. (This used to be explicit in the call -- something about multiple papers from an author being subject to more scrutiny -- but I don't see it in this year's call.)"

This year, there were a few individuals who openly ignored this implicit rule, a few had 10+ papers most of which were probably due to being a "lab director" with the student/postdoc authors...

I personally heard from some of the older and distinguished researchers in the field found this practice "distasteful" to put it mildly.

I actually find that ISIT is now too large.

One last comment, CS Theory related, I thought that the Friday Plenary by Wigderson was very weak--i.e., fluffy and not enough intellectual content, he should have included some rigorous mathematics, given some theoretical insight to the interesting problems he discussed. IMHO, Calderbank's plenary was fantastic, with just enough mathematical detail and well presented, the others were quite weak, gave the impression that they were put together quickly, and the Shannon lecture was not one of the best either.

There were some good talks mind you, I am just disappointed with the level of what are supposed to be the highlights of the conference.

Perhaps you didn't miss much by not attending :-)

Michael Mitzenmacher said...

Anonymous 4: Thanks for the conference review -- perhaps others will respond to agree or disagree!

I do think there is a concern about conferences becoming "too large". You can get lost trying to find the right room at a large ISIT conference.

I guess this leads to an interesting optimization question. What is an "optimal" conference size? [To start, what exactly are we optimizing for?]