Banazîr the Jedi Hobbit (banazir) wrote,
Banazîr the Jedi Hobbit

  • Mood:
  • Music:

Day 11: The Shoulders of Giants

UK Trip 2005: A Tronkie Travellogue
Day 11: Edinburgh, Scotland (IJCAI Conference)

If I have seen further, it is by standing on the shoulders of giants.
    -Isaac Newton

09:00 - 19:05: The third day of technical presentations at the 19th Biennial International Joint Conference on Artificial Intelligence (IJCAI-2005).

Left: Four banners outside the Edinburgh International Conference Centre (EICC).
Right: One of the escalator lobbies.

The morning's invited talk was by Daniel Wolpert of neural networks research fame, on "Probabilistic Models in Human Sensorimotor Control". Wolpert, who has moved to the University of Cambridge and established a new sensorimotor lab, is probably best known for elucidating backpropagation of error after Minsky and Papert's book, Perceptrons, but before backprop had been popularized by Rumelhart and McClelland and widely studied in the west. Shun-Ichi Amari was doing the same thing in Asia in those days, the early 1970s.

Back in the mid 1990s, Wolpert's stacked generalization technique made waves in the machine learning world, where it figured prominently in research on committee machines, including the dissertation of William Hsu, a young machine learning researcher who followed Wolpert's work.

I raised my hand, but the student volunteers were trying to be fair and give every quarter of the auditorium a chance to field a question. After the audience dispersed, I went up on stage and asked Wolpert my question: "You are looking at some reflexive motor tasks and some, such as drawing, that are at least driven by deliberation and more cognitively intensive processes spatiotemporal learning and planning. Do you see the latter as a possible way to test certain models from the NIPS/ICONIP/IJCNN community on basal ganglia that are based on looking at lesions and trying to see when and how much prediction is done?" Yes, Wolpert replied, and this is very much an open problem being addressed in some work being done at USCD. (Note to self: look this up!)

Left: The EICC auditorium.
Right: The top floor of the EICC.

Far left: A view of the street from between the second and third floors of the EICC.
Center left: One of the helpful hostesses.
Center right: The ground floor, with its many sitting areas.
Far right: The business area (conference finance desk and other service desks).

The next set of technical sessions included Temporal Probabilistic Inference, Learning Subjective Represenations, and Belief Revision. I attended two of these:

  • KVETON and Hauskrecht

  • NG, Pfeffer, and Dearden - I asked Brenda a simple question about dynamic representation and can't remember right now exactly what it was, but I remember that she gave a good answer.

Just before noon, I saw my old classmate, Vadim Bulitko, and talked briefly with him. Then I spotted Dan Gaines and talked at length with him. Dan was going to go horseback riding, but I convinced him to take the one-day Timberbush tour to Loch Ness instead.

I then go out and spend the lunch hour taking photos around town (which will appear in an album to be posted later). Here are just a few samples:

Left: The exterior of the Grosvenor Hotel. Look at that sign! Well, it beats a sentient shock of wheat...
Center: A double-decker Lothian bus off Princes Street.
Right: An ale truck arrives at an Edinburgh pub.

Oh, yeah... it's icon time...

I buy a present for Banadad and Banamum, a couple for other friends back home, and head back.

The second invited talk of the day and the sixth of the conference is by Bart Selman of Cornell, on "The Next Generation of Automated Reasoning Systems". I manage to get a question in, on technique selection, and fel quite satisfied because this one (the chief topic of my Ph.D. graduate hpguo's dissertation) was quite salient.

Left: Bart Selman, in front of the British Computer Society (BCS) Specialist Group on Artificial Intelligence (SGAI) logo - BCS-SGAI is roughly like ACM-SIGART in the USA. The IJCAI-2005 T-shirts look like this, too. So far, I haven't gotten one.
Right: The lovely IJCAI-2005 logo screen. The programs, posters, and postcards all look like this. I'd buy this on a T-shirt in a second if they made them. (The attentive reader of tanelos will notice that I even used these as the colors of House Wyvern.)

The afternoon sessions feature quite a few good topics: Ensemble Methods in Learning, Grounding Language in the World, Multi-Agent Systems, Learning 1, and Probabilistic Reasoning and Applications. In fact, so many good topics are there that I elect to chat with a postdoctoral fellow from Queen's University, Jim Davies, instead. We talk about:

  • case-based reasoning (CBR) and analogical reasoning

  • derivational analogy in software engineering

  • program synthesis

  • the importance of cognitive modeling as "the ultimate AI topic" versus application topics such as bioinformatics and data mining

  • generative models

  • classical symbolic AI and modern AI

  • the old controversies: "neat vs. scruffy" and "symbolic vs. subsymbolic" AI

Around 17:00 I happen across Eyal Amir (UIUC), Brenda Ng (Harvard), and Hei Chan (UCLA) having a chat, and join in. From 17:25 to 18:05, I explored the EICC, and talked briefly with L. Enrique Sucar, one of the RUR organizers at IJCAI-2003.

It turns out that Geoffrey Hinton has a back injury and can't travel here from Toronto to give the Research Excellence Lecture. I only heard Kaelbling talking about it this morning, and it seems it was touch and go until just the day before. (The spectacular air crash at Toronto airport didn't play a factor at all.) Rather than cancel the talk or play a recording, the conference organizers have arranged for one of his students to flip his PowerPoint slides while he gives the talk by telephone, patched over the intercom.

Hinton's talk on "Sigmoid Belief Networks" is very technical, but it takes me back to the mid-to-late 1990s when I and Jesse Reichler were enthusiastically eating this stuff up at Illinois, and giving seminars in our Artificial Neural Neworks and Computational Brain Theory (ANNCBT) group about it. As I am listening to the talk, my thoughts drift back further. Let me tell you a story.

Once upon a time, in the mid-1980s, there is a pre-teen boy who lives in Maryland. Now, this kid is fairly bright, but he has no idea what he really wants to do. He likes computers and he thinks he can program, but his math and science skills are not that sharp. The boy takes the SAT in 7th grade as part of the CTY talent search, and he does suprisingly well (600 Verbal, 650 Math), but his career options are still a cipher. The middle school guidance counselor puts him through a battery of tests and he's clearly not mathematician material. The closest job options in the book are "Actuary" and "Systems Analyst". The boy's mother worries, and during the summer of 1986, takes him to see her brother, who is recently married and in the middle of a Ph.D. program at a good computer science university, Carnegie Mellon. The young man quizzes his nephew and pronounces him adequately bright and high in logical reasoning aptitude. He had given the boy an intro CS pilot book based on a pseudo-Pascal programming called Karel, which had thus far remained mostly unread. "The boy has no patience," complains his mother. "Can you give him something more challenging?" she asks. "I'll see what I can do," replies the uncle. He rummaged through his cabinets and finds a paper. "Study this," he says, "and come back next summer, and see if you can explain any of it."

Well, as you've surely guessed, the boy was yours truly, and the uncle was Kai-Fu Lee. What you may not know is that the paper was:
Hinton, G., Sejnowski, T. and Ackley, D. (1984). Boltzmann Machines: Constraint Satisfaction Networks that Learn. Technical Report CMU-CS-84-119.

Small world, eh? One of the reasons I'm a computer scientist and an AI researcher; I kid you not.

Hinton's student later accepts the IJCAI Research Excellence award on his behalf. There is no IJCAI Computers and Thought award this time.
19:05 - 20:30: Go to a place called Shakespeare's for dinner. They have 2-for-1 specials on Mondays and Thursdays, so we get some leek and pork sausages, fish and chips, fried eggs, and other goodies.

Left: Shakespeare's, where three Chinese people got fed and watered for less than £8 courtesy of a Thursday night special. Might I recommend the Kronenberg 1664, a nice smooth lager.
Right: Hey, miyeko! What's a "Japanese Chinese Restaurant"?

20:30 - 23:30: Home for some nice telly watching. As we will be on the road tomorrow, I make sure to avail myself of Ye Olde Facilities and pack up all my stuff.

Tags: artificial neural networks, bnj, conferences, edinburgh, engineering, generative models, ijcai, lager, machine learning, motor control, neural networks, planning, research, scotland, sensorimotor control, teaching, travel, travellogue, uk

  • Post a new comment


    default userpic

    Your reply will be screened

    Your IP address will be recorded 

    When you submit the form an invisible reCAPTCHA check will be performed.
    You must follow the Privacy Policy and Google Terms of use.