AAAI 2007 at a Glance

AAAI 2007
So far, it has been a nice experience. I had the chance to talk to intelligent people like, Ronald Parr, Randall Davis, and David L. Waltz. I also met cool graduate student such as Jacob Beal, and my friend Alex Strehl. The Man Vs Machine Poker competition drew lots of attention and ended with 1 draw, 1 win, and 2 losses for the computer. Notice that human players were professional players and internationally known to the world. In terms of the papers, I liked many of them that I should select from later on, but here is potential list:

POMDPs:
Point-Based Policy Iteration
Shihao Ji, Ronald Parr, Hui Li, Xuejun Liao, Lawrence Carin

Indefinite-Horizon POMDPs with Action-Based Termination
Eric A. Hansen

Search:
Near-Optimal Search in Continuous Domains, Samuel Ieong, Nicolas Lambert, Yoav Shoham, Ronen Brafman

RL:
Nectar: Temporal Difference and Policy Search Methods for Reinforcement Learning: An Empirical Comparison, Matthew E. Taylor, Shimon Whiteson, Peter Stone

A Reinforcement Learning Algorithm with Polynomial Interaction Complexity for Only-Costly-Observable MDPs
Roy Fox, Moshe Tennenholtz

Compact Spectral Bases for Value Function Approximation Using Kronecker Factorization
Jeff Johns, Sridhar Mahadevan, Chang Wang

Efficient Structure Learning in Factored-State MDPs
Alexander L. Strehl, Carlos Diuk, Michael L. Littman

Cogntion:
Predicate Projection in a Bimodal Spatial Reasoning System
Samuel Wintermute, John E. Laird

Etc.:
PLOW: A Collaborative Task Learning Agent
James Allen, Nathanael Chambers, George Ferguson, Lucian Galescu, Hyuckchul Jung, Mary Swift, William Taysom

Recognition of Hand Drawn Chemical Diagrams
Tom Y. Ouyang, Randall Davis

Acquiring Visibly Intelligent Behavior with Example-Guided Neuroevolution
Bobby D. Bryant, Risto Miikkulainen

Victoria, you can not hide anymore!

Currently, I am sitting on one of the sits on the Queen of Victoria ferry and enjoying the beautiful scenery. You know sometime you plan in every single detail of your trip and then it gets all messed up? A while ago, I planned to go to Victoria for a one day visit. I used all of my talent in searching the internet and asked my friend in Victoria and considered all of the options which can take me from UBC campus to the Victoria island. Nowadays, we have nice online services such as trip planners. You can specify your departure location, destination, and your desired time, and the website takes care of the rest. So everything was set. Here was the plan:
* 5:30 Wake up
* 5:45 Prepare
* 6:03 Bus # 17 to Vancouver Downtown
* 6:40 Switch Bus to Airport Station
* 7:05 Switch Bus to Tsawwassen Depot
* 8:00 Boarding Ship
* 9:40 Victoria

Well seemed pretty good to me! Although ... Here is what actually happened:

* 5:40 Woke up
* 5:45 realized the mistake on the date of planning!
* 5:50 realized that there is no bus to the ferry for my desired time!!
* 5:55 Prepare!!!
* 6:05 Talking to the people on the lobby
* 6:15 Taking a cab to Central Station
* 6:30 Buying a ticket package to Victoria!
* 6:45 Boarding the bus
* 8:00 Boarding the ship
* 9:40 Victoria

It worked in the end so I can not complain! :) The view on the deck is spectacular. Small islands full of trees surrounded with water.

Contradiction

Should I feel guilty that I was half-sleep during the talk of someone like James E. Smith ? Well I did not after seeing Stuart Russell in my row with half-closed eyes!

Why parallel?

First of all, I just realized that it has been a while since I posted something here. It is a bit odd since I was not busy at all so maybe my business has a high correlation with me posting stuff :S

I am running many experiments and some of them are taking a lot of time. No, ... A LOT! you might say: "Come on you guys are in CS and should have the best computers and stuff to run programs". Well the answer might be true, but corresponding to that demands are higher as well. If you want to say my experiments are statistically significant, you should run them at least 30 times, which in my case it can translate into 7 days! I finally decided to implement a script to take advantage of many CPUs on the network at the same time to run these separate parts in parallel so I can see my results 30 times faster! It was a lot of pain and involved wrestling with Linux commands but it was finished today, thanks to the help of my lab mates! Is it working that fast? well due to my lack of luck our main cluster is down, but I am aiming to run it as soon as it is back. Wish me luck! :)