For those interested in my Honours project, I’m participating in a worldwide competition called RL Competition 2008. The participants must design a smart computer agent to ‘learn’ how to do certain actions, such as fly an RC-Helicopter or play Tetris (which is harder than you think for a machine). The winner wins a cash prize (at least that’s what I thought until further investigation where I found nothing mentioning it.) and goes to Helsinki, Finland to attend the event.
To teach the computer, a method known as Reinforcement Learning is used. To explain it as I do to my parents, think of it as, say, a rat receiving an electric shock if it goes left and cheese if it goes right. The rat will learn that it should go right more often than left. Now let the rat be a computer agent and the reward/punishment a positive/negative number. The computer will gradually, over a number of runs, learn the best actions to take to get the best possible reward for the scenario. However, to do this, it must explore the scenario to find out what is good and what is bad and then take the good path when possible.
The domain I have chosen is teaching a computer how to play Tetris effectively. This appears to be the most popular area of the competition which makes the competition that much harder, but it’ll give me incentive to be the best one at doing it. I have until early July to get a decent learning agent going to compete amongst the other agents. Thus I’m left with about 3 more months of my Honours project to do something not directly related to the competition, such as doing the documentation or further expanding the agent.
Although the Honours project is based on the RL Competition, it doesn’t matter (with regards to my degree) if I win or lose. I just need to have learnt something and become better for it and related this experience to my supervisor. Although, i would love to win and become recognised amongst the AI community.
I will be using this site to catalogue my progress and other data to the project. This information will be private to only me and anyone else with the password due to competitive secrecy but will be released once the competition is over The progress posts are now available to the public. There’s no real need to hide them, as it increases the global pool of knowledge available to all. If other competitors saw this site, they’ll only see ideas and theories.
Also to view my project proposal in a more formal presentation, see the PDF version.
