“Applied” Papers at SoCG

In his discussion of experimental papers at SODA, Suresh comments, “It is not clear that there is a paucity of experimental papers overall: ALENEX has been attracting attention and both ESA and SoCG have applied tracks.”

That's not entirely accurate. SoCG has sort of an applied track, but not really. A couple of years ago, the theoretical and applied program committees were re-merged into a single committeee with two co-chairs, which considers both types of papers together. Since then there has been serious discussion at every business meeting about switching back to a single co-chair. The last two or three SoCG proceedings do not identify which papers are supposed to be “theoretical” and which “applied”.

The applied track at SoCG was introduced in 1997 in an attempt to connect computational geometry to nearby “applied” communities—graphics, vision, robotics, solid modeling, databases, GIS, etc.—who largely thought that computational geometry was useless. Apparently, Bernard Chazelle was asked at a conference what he did. When he said he was a geometer, the reply was something like “Dear god, why? Everyone knows that computational geometry has been dead for years!” This goaded him into forming the 1996 taskforce.

In its first year, the applied track was a roaring success, at least in terms of numbers. The conference received twice as many submissions as usual, and the program committee was forced to expand the conference to four days instead of the usual three to avoid parallel sessions. (The record number of submissions may also have been influenced by the fact that SoCG was held in the French Riviera that year.)

But over the long haul, the applied track at SoCG has been a train wreck. Some of this failure stems from our community's intellectual snobbery unmitigated gall. The applied track was really introduced to bring potential customers of computational geometry to SoCG to see that, really, we're useful, honest. “You've been doing it all wrong,” we said. “Just explain your little geometric problem to us experts; we'll be able to solve it in no time, because we (unlike you) have the right tools.”

Boy, did we step in it. Our “customers” stayed away in droves. We weren't interested in what they had to say. We rejected their papers for not living up to our standards of “excellence”. At the same time, traditional computational geometers were submitting so-called “applied” papers that were pure theory, but with a made-up story in the introduction about how this problem is actually useful to real people, honest. Or so-called “experimental” papers where the experiment consisted entirely of implementing a textbook algorithm, slapping on a few obvious heuristics, running the (unavailable) code on a few toy examples, maybe comparing the heuristics to an obvious straw man to justify their existence, and crowing over a 5% performance improvement.* It was embarassing. Very quickly, people in neighboring fields just stopped submitting papers to SoCG.

I think we've learned from those mistakes. Applied (and sometimes not so applied) computational geometry papers are nor regularly published at graphics, robotics, and database conferences, following the standards of those communities rather than trying to impose our own. Computational geometry has become much more relevant. Interactions with neighboring communities have strengthened even our purely theoretical work. But this growth is not necessarily reflected in the quality of the SoCG applied track. The best “applied computational geometry” papers solve real problems of interest in some other field. So as a rule, they rightfully go to the best conferences in that field—the best graphics papers to SIGGRAPH, the best database papers to SIGMOD/PODS, and so on—and not to SoCG.

So, while I believe that experimental and applied algorithmic research is important, interesting, and increasingly common, I don't think SoCG is the right place to look for examples.

*Unfortunately, papers like this, despite being absolutely useless (if not worse), are endemic in computer science. There are very few good experimental computer science papers. This is no surprise; good experimental work in any field is incredibly hard! (In this respect, we computational geometers were copying our neighbors' bad habits.) This could also be one cause of the perceived bias at SODA against experimental papers—most experimental papers suck!

Comments

[Suresh]

But the crux of the matter is this: experimental algorithmics papers have no place in places like SIGMOD/VLDB/SIGGRAPH. You know as well as anyone how difficult it is to write a SIGGRAPH paper and what kinds of things are emphasized (experimental rigor is certainly not the main criterion). In general, in applied places, the focus is on the application, and in theory places, the focus is on the theory. There is still no natural place for papers that address the practical aspects of theoretical work in a more general setting, hence the need for ALENEX type conferences. Moreover, since such work is arguably closer to theory than to the particular applied domain (because typically one wants to look at an algorithm/set of algorithms in general), it is more natural to send such papers to a theory conf.

Basically I am distinguishing between experimental papers and applied papers (category 3 vs categories 1/2 in the Johnson survey)

[Adrien]

Hi,

Im very astonished to see a blog about Nancy. I lived in Nancy (my family live always in Nancy) and now in Paris. you photo blog is interesting : the next time I came back to Nancy, I would try to see these fishes.
don't miss the nancy jazz pulsation in October : free music in many bars.
Adrien