Side Projects

This was a side project that I carried out during my PhD so that I could learn more about reinforcement learning. Big 2 is a four-player card game of imperfect information with a relatively complicated state and action space, and so I thought it would be an interesting project to work on (the most famous successes in RL so far have come in games like Chess and Go, which are perfect information games). Big 2 is also a game that I've played a lot with friends and which I really enjoy. This ended up working much better than I expected it to, and the final agent comfortably beats me and my friends now. You can read more about it in my AAAI workshop paper, or check out the code on Github. I also made a web-app so that you can play games against the trained agent here (warning: expect to lose if you play over a long period of time...). Here are the rules of the game.