Deep Reinforcement Learning applied to the game Bubble Shooter

In 2013 Google Deepmind invented a new algorithm, deep reinforcement learning, which led to revolutionary results. The algorithm was able to play on 46 different Atari games and beat a human expert in some of the games. In this research deep reinforcement learning will be applied to the game Bubble Shooter in order to create the self-learning agent AlphaBubble. A different game state representation is used than in the original algorithm, also experiments have been conducted with bigger action spaces. The performance of AlphaBubble has clearly become better than a random Bubble Shooter agent, although it has not been able to exceed human play yet. However, AlphaBubble should be able to invent superhuman behaviour in the future.