PLClub talk today was Mike Greenberg talking about the stuff he worked on over the summer. It's assigning measure spaces to types instead of metric spaces, but still very, very relevant to what I've been doing. Factor graphs came up somehow, too, which is kinda topical because Deolalikar's recent infamous paper mentions them. I think when all's said and done they are not really that weird; not even as technical as the wikipedia page on them makes them out to be. They're just representations of a probability distribution factored into a bunch of components, each one of which is only concerned with a (ideally small) set of variables. Any probability distribution can be represented with a factor graph, if you have one huge honking factor that covers all variables --- but you hope to work with (or decide to work with, for the sake of approximation) probability distributions that have lots of little factors, like the Ising model, where the factors are just little local neighborhoods of lattice sites.
Had a meeting and it went well; we might try an NSDI submission, if some implementation details can be figured out. That deadline is super soon though. Not sure.
I was gonna go see the new Scott Pilgrim movie tonight, but I circled around the theater looking for non-sketchy places to lock my bike, and didn't succeed. I was starting to feel not super excited by biking back home from a thoroughly unfamiliar area (Columbus and Reed) in the dark anyhow (also I bet those grapes were nasty) so I think I'll go back tomorrow during the day.