Reinforcement learning in Connect 4 Game

Abstract

Reinforcement learning allows a machine or software agent to learn its behaviour based on the response of the environment. It permits machine and software package agents to mechanically confirm the perfect behaviour in a very explicit context to maximise its performance. The distinction in reinforcement learning for supervised learning is that only partial feedback is given regarding the learner’s predictions. Beside, predictions can have long-term effects by affecting the future state of the controlled system. Thus, time plays a special role. The goal of reinforcement learning is to develop efficient learning algorithms, as well as to understand the algorithm’s qualifications and limitations. It is very interesting to learn reinforcement learning from a large number of useful practical applications from artificial intelligence problems to control engineering. In this project, we focus on those algorithms of reinforcement learning. Scaling the project looks at challenges for reinforcement learning in Connect4 game, together with a review of proposed solution methods. While this list has a game-centric approach and some items are specific to the game, a large part of this overview may also provide an understanding of other types of applications.