Thursday, March 26, 2009

Parallel Session Madness

One of the last tasks I have as STOC chair is trying to organize the schedule, which involves the unpleasant subtasks of trying to group the talks into coherent sessions, and then figuring out what sessions go when, including what sessions will "compete" since we'll have two parallel sessions running. These tasks are frustrating since in some cases it's hard to find coherent subtasks, there are annoying constraints that arise (4-paper sessions should only run against other 4-paper sessions so there aren't holes; in general, an "algorithmic session" should run opposite a "complexity" session?) and decisions that people surely won't be happy with (nobody likes being opposite a "trendy" session or a "name" author, and nobody likes being placed in the last session of the day). And, of course, there aren't any tools for this (that I know of).

Last night, after already putting hours into this, it hit me -- slots should just be assigned randomly. Obviously this idea partly stems from my own frustration -- why am I spending time on a problem with no well-defined solution (and that I'll only hear complaints about afterward)? It has the advantage of making any "decisions" that people might find disagreeable an act of chance, instead of an act by me.

But I also thought of a good reason to do it this way. Categorizing and grouping talks in this way exacerbates the subdividing and splintering of theory into subgroups that know little to nothing about what everyone else is doing. Not working in property testing? Great, there's a whole session you can skip. Not so interested in computational geometry? Now you can sleep in for the morning session.

If FOCS/STOC are supposed to be flagship theory conferences each year, and they're the conferences where all these subgroups come together, maybe this session grouping is actually detrimental. Random mixing of papers could encourage people to stay in the room and see something new. As long as you're down there for the talk you wanted to see, you might stick around and see that next talk that would have been in a session you never would have gone to.

Yes, I know the price might be a few more times where two papers with a potential shared audience run at the same time. (Maybe we can still use that session information; find a random schedule that doesn't put two papers in the same nominal session at the same time.) But it seems like the payoff could outweigh this consideration.

Not that I'd really do this. I don't know of any conference that has. Although there is still time before I have to turn in the schedule. What would all of you think?

(PS -- yes, I know about the approach of having each talk given twice randomly, to help avoid conflicts -- nice in theory, not do-able in practice. :) )

25 comments:

Rasmus Pagh said...

I suppose that the following restriction on the random assignment would help eliminate most conflicts: Two papers that both received bids from some PC member should not be placed at the same time.

The conference management software would then just need to know how many parallel tracks it should schedule to suggest a random matching of the papers.

Anonymous said...

ask us to vote on similarity information (based on titles and abstracts). then do a min cost perfect matching to see who goes parallel. then k-medians clustering to get the sessions. its probably within polylog of optimal. in some metric.

Yuriy said...

What about videotaping each talk and making it available online?

This actually can be quite a useful resource for the researchers, who can then reference the video on their websites to assist the visual learners in understanding the material.

Luca said...

That's a great idea!

Suresh said...

videotaping is an excellent idea, IF you can do it without it costing too much. I was looking into doing this for SoCG, with videolectures.net, but the price was too much.

as for random assignments, why not do it ? you have the power ;)

Michael Mitzenmacher said...

Anon #2 : I'll assume that's some sort of "theory joke". :)

Yuriy : Great idea. Too expensive. (Yes, I know it shouldn't be too expensive, but keep in mind that renting a projector at a hotel for three days often costs as much as it would to buy the projector, and extrapolate from there.)

Luca : Just so we're clear, it's my idea you think is great, not one from the previous comments? (When asked, I hope to blame you when I implement it...)

Suresh : I assume since you're a regular blog-reader you're teasing me. I think we've seen earlier what sort of wackiness can ensue when a PC chair uses their power in what I would call reasonable, straightforward ways that don't match some people's preconceptions. I think this would be a bigger breach of status quo than anything I did at the PC meeting. It would, of course, also be much more fun.

Chris Okasaki said...

Anonymous #2 may or may not have been joking, but, seriously, why not open it up to public input?

After accepted papers are announced, have a week or so window where anybody can go to a website and check boxes that say "I want to attend papers X, Y, and Z".

Then randomly choose a schedule to minimize conflicts between parallel presentations for those who cared enough to vote.

If desired, you could also choose sessions to maximize the number of voters who want to see more than one paper in a session.

Pall Melsted said...

One idea is to have authors select slots in the schedule and the audience can then announce which talk they would go to for this schedule, allow authors to switch to another slot and do another round of votes from the audience. Then you just search for the Nash Equilibrium :)

Jonathan Katz said...

There are some limited cases where it makes sense to group papers in the same session: e.g., two papers on the same problem (so the second talk doesn't necessarily have to repeat all the introduction given in the first talk, and the audience avoids "context switching" and so gets more out of both talks).

Also, a lot of people attend these conferences to work with colleagues, and for those people it makes sense to have whole sessions that "they can skip".

On the other hand, the idea of encouraging people to serendipitously see a talk in another area is really intriguing...

On a different but related note, how about encouraging speakers to send their slides after the conference, and posting the slides for posterity on the conference webpage?

Michael Mitzenmacher said...

Jon --

The slides idea is very nice, and I'll suggest working with the local organizers try to get authors to send slides in for the conference web page.

Maybe there are other ways to push people to act serendipitously. Give each attendee a three random talks to attend when they check in, tell them to go to at least two... I wonder if we can find a prize for those who play.

You bring up good objections against the random schedule (and I can't really see doing it.) But the random schedule does just seem like fun to try, doesn't it... :)

Suresh said...

More ideas:

* STOC Bingo: prepare a card with secret buzzwords/slides/pics that you can only mark off if you attend the talk they show up in :). Voila, people start going to random talks :)

More seriously, the slides idea is also an excellent one, and is actually not that hard to implement. Here's what happens at conferences like VLDB. Volunteers (one per session) carry a USB key with them, and as soon as the presenter is done, they approach them and download the talk onto the key. STOC has two parallel sessions, so you only need two USB keys and two volunteers (you could even ask the session chairs, but that might involve too much transfer of the keys). USB keys are incredibly cheap.

Of course, there are probably still troglodytes who use transparencies: can't do much about that ;)

Anonymous said...

Regarding the slides: What about doing this professionally for once?

- All presenters send their slides in advance to the conference organisers.

- All slides are put on the conference web page _before_ the conference begins. The audience can have a look at the slides in advance, and use them when they are deciding which sessions to attend. More attractive slides, more audience in the session.

- The organisers provide the computer that is used to show the slides. A voluntary grad student operates the computer. The presenters are only given a simple two-button remote control so that they do not have an opportunity to embarrass themselves with their computer skills.

- Magic happens, all talks begin on time.

(Sure, everyone will complain, as they can no longer prepare their slides on the night before the talk.)

Luca said...

I think that the idea of assigning talks to sessions randomly is a great one, and since you are on record as having had it first, the blame for the consequences is all yours.

Unless you can prove, as it usually is the case, that some Russian published the same idea first in the 1950s.

Daniel Golovin said...

Actually, Blum and Raghavan considered random schedules back in 1998:
On a Theory of Computing Symposia.
They also considered the case where each talk is given twice (at random times), and show some interesting benefits of that approach.
Anyone know if any conference has tried that?

Suresh said...

Whether or not this happens at STOC, I'll definitely try the "slides in advance" idea for SoCG 2010 if I can work it out with the PC chair and steering committee

M said...

The goals of "mixing it up" and having parallel sessions are inherently at odds with one another. Randomization might make people more selective, not less; people sneaking in and out of rooms would increase, since many might decide to make only the one talk in the session that interests them, not all of them.

Given the amount of time spent on this and the seeming infeasibility of any automation, I wonder if a collaborative approach would be possible. Given the right set of initial rules (including full transparency with regards to who's modifying what), a wiki might be a way of configuring the fine-grained level. This is not as ridiculous as it sounds if it's done as part of a system with the suggestions here, such as Chris Okasaki's voting suggestion. With or without the wiki, that suggestion would also benefit from some categorization on the part of the authors (which is often asked for anyway). In a generalization of the problem with more than two tracks, there might not be enough data based on audience voting alone, especially if you require the audience to register before voting. How many people are going to have the time to both register and vote early?

As far as encouraging people to go outside their comfort zone by scheduling dissimilar talks in the same session, that already happens when the scheduler is unfamiliar with the topics he or she is scheduling! It also happens when a talk needs to be shoehorned in somewhere due to the need for four-talk chunks, when a talk doesn't fit in neatly to any given category, and, presumably, happens when colleagues working on different problems get roped into going to a presentation on a topic they'd never discuss otherwise.

Along the lines of making slides available would be to add optional informal online abstracts to the formal published ones. Those who choose to give the optional abstracts would try to give ones which assume no previous knowledge of the topic, encouraging the uninitiated to learn more. Hopefully that would also indicate which talks would be more likely not to lose outsiders in the first two minutes, too.

Anonymous said...

"(Sure, everyone will complain, as they can no longer prepare their slides on the night before the talk.)"

Who prepares their slides the night before? I've always gotten them done during the talk before. ;)

Yuriy said...

I'd like to contest the claim that it's too expensive to record and post videos of the talks. Yes, it's certainly expensive if you go through the proper channels (e.g., a professional company or the hotel); however, what if you buy 2 digital cameras (about $300 each + $100 for "tapes"), get student volunteers to record each talk, and post these using a free service (e.g., Google video). It will take some precious time to organize, but following this "low-tech" approach would actually be rather cheap (+ produce 2 cameras in the end). Alternatively, one could rent cameras from rent-a-center for under $100 total.

If the economic times were better (i.e., if I, a non-author, could afford to attend STOC), I'd volunteer to organize it. Perhaps someone else could?

Anonymous said...

Given how much value we would get out of the recorded talks, it would have to cost a lot to be "too expensive".

Michael Mitzenmacher said...

Yuriy + Anon 19:

I have to admit, I'm skeptical that videos on a digital camera posted to something like Youtube is going to be good enough to offer a lot of value that Anon 19 seems to think we'll get out of recording the talks. Professional videos could offer significant value -- but at significant cost. But that's just my opinion.

There's also a fair amount of administration that would have to be set up; I'm sure we'd need an appropriate legal form signed to record the speaker. So I doubt it's going to happen this year. It would be an interesting experiment for the future.

I do have a question, though. If the talks are so amazingly valuable, why is it (it seems to me) so many people skip so many talks?

Anonymous said...

"There's also a fair amount of administration that would have to be set up; I'm sure we'd need an appropriate legal form signed to record the speaker."

Certainly not. Nobody else ever does.

Michael Mitzenmacher said...

Anon 21:

Almost all the talks I've given where I've been recorded, I had to sign an official form saying that they had my permission to record and distribute the talk. So I don't know what you mean by "Nobody else every does." In my experience, essentially everyone does.

Since the distribution of the talks would legally be under the IEEE or ACM as the official sponsors of the conference, I'm pretty sure they'd insist on a form.

In short, I think you're wrong, but would be happy to be made aware of other information.

Ragib Hasan said...

In reference to your reply (#22) to anon(#21), I think most of the conferences do not require signing forms. I gave a talk at USENIX FAST 09 last month, and all the talks there were videotaped, but none of the speakers had to sign any release forms. I also recall that ACSAC had some sort of recording (at least audio), but didn't ask me to sign a form. So, perhaps signing a release is not a prevalent or required practice.

Mark Knell said...

If you've ever attended a conference in Open Space format, it's eyeopening.
http://en.wikipedia.org/wiki/Open_Space_Technology

The classic approach has no pre-set agenda whatsoever and relies on spontaneous self-organization of both content and timeslots. The slots themselves are fixed, but are unassigned until the conference actually begins. In an inaugural meeting, presenters choose slots round-robin. The gating factor is often a microphone at which they must announce/pitch their topic before selecting a slot.

I've been to three such conferences, all related to the agile programming movement. The technique works surprisingly well there--both in terms of efficiency and content--though it's hard to say whether that's an artifact of the agilistas themselves, who of course are in love with the format.

Anonymous said...

You don't even need to record the video, the slides and the voice will be great. And recording with a digital camera is not that hard. Stanford has done this for a series of lectures (the first one by Knuth if I remember correctly) and they are great.

About getting the signatures, I think you may need it if you want to release it officially, but if you are just going to post it on youtube a simple notification to speakers beforehand to allow them to opt out if they want will be enough IMHO.