Machine Learning with the Pong Game: A Case Study

Being one of the earliest computer games, the Pong game is well-known for its simplicity, which makes it suitable for becoming one of the very first problems in Artificial Intelligence and Machine Learning: The goal is to create a self-playing agent that can compete against humans. In the past there have been introduced various Machine Learning approaches to solve this problem. This paper gives a summary of some notable techniques to creating a self-learning agent for the Pong game. In addition, it proposes a template for developing this idea into a full-fledged application. An implementation in Java is available online.