CIG 2011

This is a little late coming, but I was recently in Seoul attending the IEEE 2011 Conference on Computational Intelligence and Games (CIG ’11). This is a conference I have been wanting to attend for quite some time, as it is basically right up my (research) alley.

Not only was I attending the conference, but also presenting my paper: ‘Using the Online Cross-Entropy Method to Learn Relational Policies for Playing Different Games’. The presentation went reasonably well, and I was on time. The range of questions received were mainly concerning the extensibility and comprehensibility of the relational policy, behavioural cloning behaviour into a policy (something which may need to be done for more useful module learning), and Turing behaviour, which was quite surprising.

The question about the Turing behaviour regarded a clip of the agent in the Mario environment, where some of the audience believed that the agent was a close fit to a human and could be entered into the Turing track of the Mario competition. This has never come up before and was an unexpected side effect of the agent’s learning and action resolution behaviour. Entering the Turing track in the next Mario competition is practically free, so I might as well enter next time it comes up.

Unfortunately the Ms. PacMan agent entered into the PacMan vs. Ghosts competition performed very poorly. I was fairly sure it wasn’t going to do so well when I submitted it as there was a very large discrepancy between the scores of the experiments and the scores of the agent when run by individually. I am still unsure why this error occurred, but it will be remedied. Next year I will do better!

The rest of the conference was very enjoyable, with a large emphasis on evolutionary methods and procedural generation for the various tasks. There was very little presence of traditional machine learning algorithms, which may be simply due to the nature of the problems, or just because evolutionary methods are more fun.

Finally, Seoul the city was excellent. The food was delicious and cheap (if you went to the right places), and the city is quite clean and relatively uncrowded. The worst part was the heat, which was 100% humidity all the time. I managed to extend my stay by 3 days to look around the city after the conference which was well worth it.