Wednesday, May 24, 2006

Going to STOC and FOCS

With the exception of STOC in Montreal, where I could not go for a visa problem, I have been to every STOC and every FOCS since STOC'97. The official (and historic) purpose of scientific conferences has something to do with the rapid dissemination of research results. In reality, most papers presented at STOC/FOCS had long been disseminated before the conference takes place. There are many other reasons, however, why I like going to these conferences and why I think they are important.

For one thing, they keep our community together. Over time, the scope of theoretical computer science increases, as we become interested in more and more subjects, like error-correcting codes, quantum computing, game theory, and so on. I think it is very important that we have a single place to go to, where results on all these subjects are talked about, and where people can notice connections between what they do and the new things they hear about. Avi has already discussed very eloquently the need to keep in touch with what other theoreticians are doing, so I will refrain from repeating his points. I just want to say: can we have single sessions in STOC too?

This STOC in Seattle had one of the strongest programs among conferences that I can rememeber. There were perhaps four papers each of which would have easily been considered the best paper at some earlier FOCS/STOC. Indeed, the quality of most papers was outstanding.

Most talks were very good too. James Lee says that the reason he goes to talks is that often the speaker will say one sentence that gives some insight that would have been very hard to extract from the paper. That sentence, he says, is what makes the talk worthwhile. He is quite right, and, indeed, someone who went to several talks and who writes often on the web should collect these sentences and put them online. The 20-minutes format helps. It is very hard to prepare such a short talk, but the time constraint puts some pressure to cut to the chase and just say what the paper is about and what new ideas are there in the proof. Among several other talks that I enjoyed (such as Anup Rao's and Irit Dinur's), I liked Guy Kindler's talk on this very technical result. I can't say how well he explained the problem, because I was already familiar with it, but I think his explanation was very clear. With a few minnutes to spare, he started to talk about their very technical proof and he gave an explanation for "why" the result is true that was very clear and that I had not gotten from reading the paper and thinking about related questions.

Normally, another reason why I enjoy going to FOCS/STOC is to see my far-away friends that I get to meet only once in a while, and to catch up with what is new in life and in theory. In this conference, however, I went to talks in almost all the sessions, except when jet lag made me oversleep. (Hence I missed the session on zero knowledge and Russell Impagliazzo's invited talk, which were highly praised by those who attended.) In fact, a few times, there were two talks that I was interested in that were scheduled against each other. Bad parallel sessions, bad, bad, parallel sessions!

By the way, I am aware of a problem here. If there are single sessions then fewer papers can be accepted, and if I thought that almost all papers were so good, how can I support a system that would have led to the rejection of some of them? Well, by now my readers know that coherence is not the strong point of my posts. I don't know, maybe we should have a four-day conference. Or maybe the good rejected papers would be resubmitted to FOCS in Berkeley. Perhaps, by random fluctuations, the quality of the other submittd papers will not be so high, and things will even out.

Which reminds me: people who attended the business meeting at this STOC and at the previous FOCS might have some doubts, but I hear that, yes, there will really be a FOCS'06 in Berkeley.


  1. Blogger Scott
    5/24/2006 08:54:00 PM

    I'm glad you liked the technical program -- we on the PC were lucky to get such outstanding papers.

    Regarding parallel sessions, what do you think of what was done at STOC'04 (ask people ahead of time which talks they want to attend, then use a matching algorithm to minimize conflicts)? I saw firsthand that scheduling ~80 talks is a very hard optimization problem.

  2. Blogger Luca
    5/25/2006 04:35:00 PM

    The scheduling was good. Compared to other conferences, I attended more talks and I had fewer conflicts. But the main problem I have with parallel sessions is not the obvious conflicts. It is the fact that it is harder to just wander into a talk in a different area that someone recommended to you.


Post a Comment

<< Home