MCTS with influence map for general video game playing

In the General Video Game-AI competition in 2014 IEEE Computational Intelligence in Games, Monte Carlo Tree Search (MCTS) outperformed other alternatives. Interestingly, the sample MCTS ranked in the third place. However, MCTS was not always perfect in this problem. For example, it cannot explore enough search space of video games because of time constraints. As a result, if the AI player receives only limited rewards from game environments, it is likely to lose the way and moves almost randomly. In this paper, we propose to use influence map (IM), a numerical representation of influence on the game map, to find a road to rewards over the horizon. We reported average winning ratio improvement over alternatives and successful/unsuccessful cases of our algorithm.