Chow a bowl of hot oatmeal and head to the CAB (CS building) for the first of many classes I’m sitting in, Reliable and Interpretable AI. The class looks really cool, we’re going to learn how to prove seemingly difficult things about neural nets and learn how to learn human-interpretable models of various forms. I meet a cool dude named Martin who’s fascinated by trying to reconcile machine learning with any kind of reasonable epistemology. Charming fellow, very Johanesque, and with a big handlebar mustache.
Mittagsseminar
RIght after that is the Mittagsseminar in Theoretical Computer Science, a little half-hour presentation that happens every Tuesday and Thursday! 30 minutes is not long enough for the talks–55 minutes at the UW would be more sane–so the speaker hauls at an unbelievable pace through an interesting math puzzle about a dude with robots, some of whom are faulty, trying to find the gate along an infinite fence.
Expectations are King
After that, probabilistic algorithms! Some examples which are very cool but I’ve seen the basic idea before: take an expectation of a count and if it is less than 1, infer that in some situations the count must be zero. The new part is awesome though–using that idea to construct a binary search algorithm for a satisfying assignment to a Boolean circuit (under certain conditions, otherwise \(P = NP\)) After that exercises for the same class, then head home to Culmann, make myself a tasty ham sandwich. Everything else is negligible!