Daaaagstuuuuuhl!
I’m happy ensconced once again at the Schloss Dagstuhl International Conference Research Center for Computer Science in Germany for a week, and it’s swell.
The short version is that Dagstuhl is a 1700’s manor house converted into a dedicated CSci research facility. They run a series of weekly workshops on different topics year round. Every two years they’ve had one in January or February on the Theory of Evolutionary Algorithms, and I’ve had the pleasure of coming three times (including this visit). It’s a small-ish group of about 50 cool people focusing pretty intensely on the subject, and it really stretches my head in cool (if sometimes painful) ways.
I could talk a lot, but I should be doing research, so I’ll leave you to look at the pictures and read the history. (Getting ready to come here in the midst of classes and admin duties is largely the reason for not much blogging recently, and it’s likely to remain quiet as a result.)
This morning’s round were on co-evolution and had some very nice material.
- Ken De Jong gave a nice talk on work by his student Popovici that, among other things, showed how you could use best-reponse curves to understand the dynamics of co-evolutionary systems with some nice results on unstable and even chaotic dynamics.
- Edwin De Jong (no relation) talked about his work on underlying objectives in multi-objective optimization, which was very cool since one of his papers had been the primary focus of one of my Senior Seminar students (Jon Q) last semester. Jon and I had really enjoyed that paper and I’d gotten a lot of ideas from it, so I’m looking forward to talking further with Edwin about his work.
- Paul Weigand suggested that one of the things that co-evolutionary systems “did” (or were “good at”) was to evolve robust solutions as much as evolve (close to) optimal solutions. While there are lots of issues (which Paul raised and acknowledges) about what “robust” exactly means and how all this depends on problems and representations, the basic idea seems to make some intuitive sense. Any process that has to act in a noisy or stochastic environment needs to be robust to a degree against that noise, which makes it seem plausible that (given the correct representation and operators) evolution would be fairly good at generating solutions that are robust to other forms of change.
Soon it’s tea and cake, and then we’re back in for the afternoon session! Huzzah!
