Visualizing the Concepts of Machine Learning Techniques Through the Game of Tetris

JOSIAH M. BELL

Complex concepts in the field of Computer Science can be difficult to understand for those outside of the field. Of these concepts, artificial intelligence and machine learning are among the most difficult for non-computer scientists to comprehend. One of the best ways to showcase these concepts is through video games using artificial players. These artificial players are artificial intelligences that learn a game’s mechanics and strategy through a machine learning algorithm customized for the game it is learning. Creating a visual representation of an AI playing a game makes it easier for anyone to grasp
The purpose of this project is to show and compare the visual representation of two machine learning techniques through the puzzle-style video game Tetris. The two machine learning methodologies showcased will be Genetic Learning and Reinforcement Learning. The game will be a clone of Tetris that will host the learning algorithms. It will show the process of each algorithm as it trains the AIs, as well as display a live visual of the trained artificial intelligences performances. The two machine learning techniques will be broken down and explained, and the results of each algorithm in training the artificial intelligences will be displayed and examined. The results that will be shown will be comparisons of the two algorithms including, the pace of each algorithm’s training and the respective performance of each artificial intelligence according to the amount of training received, the performance of each algorithm’s optimally trained artificial intelligence, and each algorithm’s ability to indefinitely perform in Tetris. These results will be displayed using graphs and visual representations. Some interesting issues in the project include developing an environment of Tetris that each machine learning algorithm can use to learn, discerning which parameters are necessary for the algorithm to use to learn the game’s mechanics and strategy, and how best to collect the training data and respective performance and translate it to
visualize the comparison results. By the end of the project, both machine learning techniques will have been dissected, explained, compared, and one of the learning techniques will be chosen as the most effective in learning and playing the game of Tetris. Through this, just about anyone will be able to understand the basic concepts of machine learning and visualize these concepts through something as simple and relatable as the game of Tetris, as well as get to see the differences between just two of the many machine learning techniques.