My response is linked there as a PDF: I thought I'd place the main text here, just to solicit some comments. After writing this, I've done some digging for data (which will hopefully appear in a post shortly) that brings up an interesting angle to the 'accept more papers' discussion.
My response: This is a terrible way of solving the problem, because...There are better solutions!
It is puzzling to me that we're jumping straight to an alien (to us) conference model, when there are proven hybrid conference models that exist within the larger (conference-driven) computer science community. ICALP (theory), SIGMOD, VLDB and ICDE (the top three database conferences), ICML and NIPS (the top machine learning conferences), KDD and SDM (the main data mining conferences), SOSP (the main OS conference), and (to a degree) SIGCOMM and INFOCOMM (networking) all have the following model:
- A main research ``track'', with peer reviewed papers. Optionally, an ``industrial'' track for more applied work.
- A set of workshops, for which topics are solicited via a call for proposals.
- A set of tutorials, for which again topics are solicited via 5-page abstracts and reviewed. Panel discussions and demos (optionally)
Other variants include limiting the number of actual talks, while still including all accepted papers in the proceedings (the remainder of the papers are presented as posters). This opens up more time during the day for satellite events as well.
Note: I've begun noticing more tutorial events at STOC (STOC 2010 has a full day of these). This definitely is progress, although I believe that workshops draw more attendance. I also think it's important to solicit these from the community, rather than having the PC decide topics. Doing so both increases participation and increases the sense that the community is part of the whole conference.
Math meetings are fractured
I've never attended an AMS meeting myself, but from all accounts (and I have heard MANY), they are highly fractured. The meetings draw attendees from the spectrum of mathematical areas, but they all essentially group together in tiny miniconferences - I've heard numerous stories of multiple rooms in which the only people listening to a talk are the five other speakers and a few graduate students. This is not the way I want to see our flagship theory conference evolve.
People won't come
For better or for worse, we're used to the 'publish-attend-present' model of CS conferences, rather than the 'meet-greet-discuss' model of math/science conferences. I suspect that many people will not come to a STOC in which there is no proceedings and no real publication attached to the presentation (at least none that can go on a CV). There's nothing wrong with the model per se, but it's not something our community is comfortable with, and since other theory conferences won't go this route, I don't see how we'll ever get comfortable with it.
I applaud the idea of shaking things up in regard to the format for STOC. I just feel very strongly that we should learn from other models that exist within the computer science community, rather than making a radical shift towards a
model that's part of a very different framework for how the publishing/dissemination process works. I am unconvinced that the proposed model would solve the problem of attendance, and it has a good chance of making STOC entirely irrelevant.