Suicidal Tendencies

An anonymous comment on my last post raises some important points:

I've heard people joke about the "suicidal tendencies" of the computational geometry community. These tendencies include:

(a) small SoCG acceptance numbers. Students are discouraged by high rejection rates and people dabbing in the field are shunned away. [Contrary to what Jeff says this does not make the numbers any harder. The moment the word is out that SoCG is no longer a long shot you-better-know-someone-in-the-PC conference, the number of submissions would go up.]

(b) no clear second/third tier conference (contrast this with e.g. complexity with CCC/STACS/MFCS or algorithms with SODA/ESA/(WADS/SWAT))

(c) plenty of weakly refereed conferences which are given very little weight by grant review panels such as the European Workshop, the Japan Workshop and CCCG.

(d) weak links to direct application areas such as computer graphics, GIS, robotics, computer vision, protein folding, computational chemistry, etc.

(e) low submission rates to FOCS/STOC/SODA (as noted by Jeff).

No one is suggesting the standard be lowered. There were 65 papers worthy of publication, only 40 appeared. Is this a desirable situation for anyone?

All these criticisms are well-deserved.

I've only ever heard two reasonable arguments against accepting more papers to SOCG. One is that higher acceptance rates devalue conference papers in the eyes of hiring, promotions, and grant review committees. They shouldn't, but they do. When I came up for tenure last year, I was asked explicitly to list the acceptance rates of every conference where I had a paper. When I've served on hiring committees, my cohorts have asked me about the selectivity of theory conferences. (SODA papers are at a disdvantage here.) And as the anonymous commenter points out, papers at low-filter conferences like CCCG are consistently ignored by grant panels. Nobody in their right mind really believes that acceptance rates are an objective measure of the quality of a conference—certainly not the quality of an individual paper—but academics have a nasty habit of inappropriately distilling qualitative things down to numbers.

The other argument against accepting more papers is that the SOCG community likes having a small conference. We like knowing that we can see all of the best results in our field, without worrying that two interesting talks will be scheduled on top of each other. We like being a small close-knit community.

That's not to say that we keep our doors closed to new people; as a graduate student, I felt incredibly lucky to have stumbled on such an open and welcoming community. But as a group, we have some unfortuantely narrow ideas about what constitutes good computational geometry. Our small close-knit community is all too easily seen as a narrow clique of elitist snobs. Witness, for example, the creation of CCCG as an outlet for frustrated Canadians slighted by the Sharir hegemony, the utter failure of the so-called "applied track" at SoCG, or the nearly complete lack of interest in important geometric developments like metric embeddings. Or the commenter's perception that SOCG is a ”long shot you-better-know-someone-in-the-PC conference”.

And yet we SOCG folks have the temerity to criticize STOC and FOCS for their snobbery. (Yes, of course! We're the ones being snubbed!) Mr. Pot, this is Ms. Kettle. I believe you already know each other.

Suppose we increased the acceptance rate at SOCG, say by adding parallel sessions. The question we have to ask is whether the resulting influx of new papers (and new authors) offset the (perhaps temporary) loss of prestige and (yes) quality? Would expanding the community be healithier in the long run, despite making it harder to stay as close as we are now?

Personally, I think it would.

No clear second- or third-tier conference? That's not entirely true. SODA is another first-tier conference that consistently publishes a significant number of geometry papers. (We don't submit much to STOC and FOCS, but SODA is another story entirely.) Geometry is also well-represented at various "lower-tier" algorithms conferences: ESA, SWAT/WADS, ISAAC, LATIN, FSTTCS, STACS, COCOON, etc, etc. But it is true that among purely computational geometry conferences and workshops, only SOCG has a nontrivial review process. Almost every paper submitted to CCCG, EuroCG, JCGCG, of the Fall Workshop is accepted.

There's an apparent contradiction here. On the one hand, the acceptance rates at SOCG are too low. On the other hand, the acceptance rates at the other computational geometry workshops are too high. (Dare I suggest that the first is a direct cause of the second?) I think what the commenter is suggesting (and I agree) is that we need something more in the middle, perhaps a more diverse range of conferences.

I'll talk about the commenter's other points in a future post; this one is already too long. For some interesting related discussion, see the recent posts (and their comments) by Lance and Sariel.

Comments

--- One is that higher acceptance rates devalue conference papers in the eyes of hiring, promotions, and grant review committees.

The point is that the acceptance ratio wouldn't be any higher even if more papers were accepted, paradoxical as it may sound.

Viz. for the last 5 years most of my papers were published in theory conferences with a 25-35% acceptance ratio even though the conference list included everything from minor conferences all the way to SODA, STOC and FOCS. What was at play here was self-selection. If SoCG were to publish twice as many papers, the acceptance ratio wouldn't double. In fact it would only move up by 3-5% points in the long term.

--- The other argument against accepting more papers is that the SOCG community likes having a small conference. We like knowing that we can see all of the best results in our field, without worrying that two interesting talks will be scheduled on top of each other.

This concept of "not missing the talk" is a logical fallacy. Parallel sessions or not, you missed 40 talks. Which 40 talks? the ones from those 40 papers that were good but got rejected.

[JeffE]

The low acceptance rate problem is ubiquitous across all theory conferences, not just SoCG. SODA's acceptance rate shot up dramatically when (quadruple!) parallel sessions were first introduced, but then settled back down as submissions went up AND acceptances went down. (The SODA accetpance rate is still higher than either STOC or SOCG, and that *does* devalue SODA papers.) Yes, the long term effect would be small (assuming past performance is indicative of future results), but we'd have to agree to sacrifice a year or two in the short run.

“This concept of "not missing the talk" is a logical fallacy.” — Agreed, but this isn't a logical argument! People are comfortable having things they way they've always been, and we make up all sorts of rationalizations to justify keeping them that way. We're not brains on sticks. Keeping a one-track conference is more comfortable than changing to a conference with parallel tracks, if only because that's what we're used to.

I don't think either of these arguments justifies keeping the SoCG acceptance rate as low as it is, but I do think it's important to realize that they have power. If we want to change things, we have to convince people that the change is in their long-term PERSONAL best interest.

--- SODA's acceptance rate shot up dramatically when (quadruple!) parallel sessions were first introduced, but then settled back down as submissions went up AND acceptances went down.

If I remember correctly this also had a bit to do with some unplanned aspects of the size increase, aspects which presumably could be avoided by SoCG if the increase is done in an orderly fashion.

All this talk about what is best for tenure, what is best for hiring, and what will disturb the delicate scientist who does not want to face the horror of having to choose between two talks. How about asking what is best for Science? What will encourage the maximum amount of innovation?

--All this talk about what is best for tenure, what is best for hiring, and what will disturb the delicate scientist who does not want to face the horror of having to choose between two talks. How about asking what is best for Science? What will encourage the maximum amount of innovation?

When trying to understand the politics of Computational Geometers (SoCG guys and the like), it's best to imagine them as Sociologists. Now, do you think most sociologists worry about the advancement of science?

[JeffE]

“How about asking what is best for Science?” — Well, sure. Why do you think I'm raising the issue at all?

”Now, do you think most sociologists worry about the advancement of science?” — Umm... YES.