Day 5: Edinburgh, Scotland (UAI Conference)
The view from my window, around 04:40 GMT Fri 29 Jul 2005.
08:15 - 08:45: Get up and dressed, drink some milk and have a quick bite, and pack for the conference.
08:45 - 08:56: Walk up to the Appleton Tower under the Potterow underpass. It's overcast and slightly drizzly again today.
08:56 - 09:32: Read and answer a few mails. I see that nanaki117 has finished the SUROP PHPnuke-based CMS, and send him a thank-you note using Hotmail. Then I get up to the lecture hall.
09:32 - 10:50: Session 5, on Applications. Fun! I always love this track; it's something to look forward to throughout the conference.
- "Robotic Mapping with Polygonal Random Fields", Paskin and Thrun. I came in just a few minutes before this finished, but I really do need to read this, along with about 3-4 years of SLAM work.
- "Efficient Test Selection in Active Diagnosis via Entropy Approximation", Zheng, Rish and Beygelzimer. Exciting title, very technical paper and talk. Presentations such as this one remind me first how technically deep UAI's expectations are and also that I need to improve my CS writing.
- "Bayes' Bluff: Opponent Modelling in Poker", Southey, Bowling, Larson, Piccione, Burch, Billings and Rayner. The Alberta folks always look like they are having such fun with their computer poker papers. I first saw one of these at GECCO-2000; there was one or two at Edmonton in 2002, and several in Acapulco in 2003. One of them, on which Rob Holte was a co-author, won best paper at IJCAI-2003. (Rob, in turn, was my faculty mentor for the AAAI-1997 Doctoral Consortium.)
- "Prediction, Expectation, and Surprise: Methods, Designs, and Study of a Deployed Traffic Forecasting Service", Horvitz, Apacible, Sarin and Liao. Wow, 3000 smart phone users just within the Microsoft Corporation. I had opportunity to drive someone's Beemer and Mercedes SUVs around Bellevue and Redmond back in 2001, and though traffic wasn't horrible, it's not typical of what you see anywhere in the Midwest save in KC and Chicago.
10:50 - 11:20: Break. Take two cups of coffee.
- Say hello to Russ Almond, creator of the Software for Belief Networks page, and talk to him about his UAI Wiki and my idea of a bayesnets wiki to be called GMWiki. He's understandably concerned about fracture of the body of material. I might let GMWiki be subsumed into the UAI Wiki or its content reassimilated cf. Memory Alpha and StarWarsWiki, but I'd like to go with the same idea. I've been cranking on a graphical models glossary since I was sitting in Sebastian Thrun's talk at UAI-2002, and very little has happened in the meantime, wikiwise, so it makes sense for me to go ahead with it.
- Ask David MacKay some follow-up questions about Dasher: Can it be used as a statistical MT tool? Can it be used for efficient intermodal communication? (I wasn't really thinking of Braille or speech recognition, but rather something really out there such as remapping sound to images, colors, tactile feedback - a sort of "Helen Keller teletype", though I wasn't sure how to express the idea. In many SF stories, too, aliens communicate moods using pheromones, and so I was thinking of a speech/gesture-to-smell dictation unit. I thought that might be a little too weird a subject to broach during a coffee break! Then again, perhaps I should be so concerned dropping wild ideas before people who invent breath mice and other such nifty input devices.) What about Pinyin? It turns out that a Pinyin version is in the works and is due out next week.
11:20 - 12:10: Session 6, Approximate Inference.
- "The DLR Hierarchy of Approximate Inference", Rosen-Zvi, Jordan and Yuille. The GMWiki needs a benchmark section and one on energy measures (Bethe free energy, Kikuchi free energy, etc.)
- "Toward Practical N2 Monte Carlo: the Marginal Particle Filter", Klaas, de Freitas and Doucet. This work was masterful, and I really need to get more active in particle filtering.
12:10 - 13:10: An invited talk, "Uncertainty and Algorithms in Structural Molecular Biology and Proteomics", by Bruce R. Donald of Dartmouth.
13:10 - 14:30: Lunch period. Snap a few photos of the inside of Appleton Tower.
14:30 - 15:30: Invited talk, "Models and Games for All the World's Information", by Peter Norvig of Google
15:30 - 16:50: Session 7, Learning.
- "Learning to map Sentences to Logical Form: Structured Classification with Probabilistic Categorical Grammars", Zettlemoyer and Collins. It's certainly interesting, though I can't really say it's fantastically practicable yet.
- "A Conditional Random Field For Discriminatively-Trained Finite-State String Edit Distance", McCallum, Bellare and Pereira. Sometimes the most successful and powerful ideas are algorithmically simple, as in this gas.
- "Discovery of Non-Gaussian Linear Causal Models Using ICA", Shimizu, Hyvärinen, Kano and Hoyer. Good paper, but... is it considered attention-whoring to use the acronym LINGAM for one's system?
16:50 - 17:00: Show BNJ v3.1 to Judy Goldsmith. She says she might be interested in trying it in the fall and will definitely be doing so in the spring.
Thus ends UAI-2005.
Outside Appleton Tower on the University of Edinburgh campus.
17:00 - 18:30: Go back to the Scottish Youth Hostel for the last time; we have checked out, but the luggage is in a free storage room. Help the folks get us all moved to the City Centre Guest House, a hostel at Shandwick Place next to Lothian Road. They charge me the pre-Festival rate, which is £504 ($942) instead of £635 ($1187). I am win!
Left: The City Centre Guest House from the outside, across Shandwick.
18:30 - 19:30: Drowse a bit. Wake up, and find that the Banafolks have gone out to get some food. Get up and take some photos.
Left: Bunks and telly in our room.
Center: The central stairwell.
Right: Another view of the bunks.
19:30 - 20:30: Dinner consists of curry cup noodles and some croissants filled with sweet cinnamon applesauce, from the Sommerfield grocery next door. I swear, this is the most convenient hostel I've ever seen. The bathroom is shared among four rooms on the floor, but there is a small kitchenette with a sink, a coffee maker, and lots of free coffee and tea. Not bad.
20:30 - 21:00: Discover that the hostel has wireless internet access! W00t! Accept an invitation into the CRG chat room from masteralida. Log into
#teuncbriefly. Nobody is there when I arrive.
21:00 - 01:00: Watch BBC news for a bit, type some notes, read LJ, and do some web browsing. Notice that nobuddy69 and gondhir have arrived in
#teunc. The conversation languishes until around 23:00, when Tripitaka arrives. We start chatting and I apprise them of the fact that we have had mostly Chinese food here in the Scottish lowlands, then foist some of my photographs of ai4din3bao3 on them. Brush my teeth and get ready for sleep. Bid 3p good night at 00:51.
01:00: Hibernate the computer and go to sleep. The bedsprings are rather painfully stiff and exposed. Oh, well; you can't have everything.