Tuesday, January 22, 2008

SODA 2008: Summaries

I've written before about the perils of blogging at the conference: another problem is that if you don't blog, people come by and start asking you when the posts are going to appear.

Before you read further, do check out posts by DavidE (II), Michael Mitzenmacher (I, II), and a "stranger in a strange land" take by Hal Daume, my colleague at Utah and an ML/NLP person whose blog you should definitely follow if you're interested in such topics.

There's been a lot of work on clustering in high dimensional spaces, and I want to say a lot more about this, especially the "abstract framework" results of Kumar, Sabherwal and Sen, but that will have to wait for another post. For now, I want to highlight the paper, "Clustering for Metric and Non-Metric distance measures" by Ackermann, Blomer and Sohler. Essentially what they do is take the framework for approximate linear time clustering in high dimensions, and remove the need for either symmetry or triangle inequality in the distance measure, replacing it by a set of sampling conditions that have a more statistical feel to them. Doing this allows them to cluster with more general distance functions, including the relative entropy or Kullback-Leibler divergence. I'm a sucker for results that illustrate "what's going on", especially in the realm of approximate high-D geometry, and this is a great example of attempting to get to the core strucutural properties needed to make an algorithm work.

Outtakes:
  • After many years, I actually missed the business meeting this time, so no drinking games or updates. From what I hear, Austin is the first choice for the venue in 2010 (we're in NYC for 2009). Claire Mathieu is the PC chair for SODA 2009.
  • Overall, I really enjoyed SODA this year: somehow I felt that there were far too many interesting papers (too many for me to attend the talks, that is), and the papers were of really high quality.

2 comments:

  1. I think the most popular choice for SODA 2010 was Las Vegas but someone claimed this might be too expensive (?). So the second choice, Austin, might prevail.

    ReplyDelete
  2. I think, in the final count Vegas got 37 votes and Austin 35, and David J. said that SIAM and SODA steering committee (him and P. Tetali, I presume) will make the final decision (mostly likely out of these two, but who knows).

    I checked the prices in Vegas for the coming weekend (Sat-Tue) this year and there are plenty of good hotels (many seem to be nicer than the one in SFO this year) with the reasonable prices (for example, New York-New York for $110 and Tropicana [where we had two STOC conference, though not everybody liked the hotel] for $55!). All these prices are well below next year SODA in New York. Of course, SIAM must negotiate the full package (incl. lecture rooms, etc), and now it's the end of January and they might prefer early January, but in any case, I don't see why it should be expensive in Vegas in January.

    I don't know Austin well, but large and good hotels seem to be not cheaper there than in Vegas. (Of course, SIAM might be able to negotiate better prices.)

    Finally: popular choice was to have only electronic versions of papers for the next year SODA + a booklet with abstracts. So, perhaps we won't see hard copy proceedings next year!

    ReplyDelete

Disqus for The Geomblog