Learning state-action features for general game playing
Dennis J. N. J. Soemers · 2023
Throughout most of my life, I have played games and-time permitting-spent some time studying or working.Already before I started following the Knowledge Engineering (B.Sc.) program at Maastricht University's Department of Knowledge Engineering, I also became interested in scripting and programming to modify the video games I played.While many different aspects of Artificial Intelligence (AI) research fascinated me, one of my favourite areas became the intersection of AI and games.The research for my bachelor's thesis focused on the development of an intelligent agent that could automatically play the videogame Star-Craft: Brood War.Later, for my master's thesis, I developed a program that could only play arcade-style video games (inspired by even older games than StarCraft),but a larger number of different ones.By this time, the department had also been renamed to the Department of Data Science and Knowledge Engineering.After a detour, where I spent time on a highly interesting non-game project at the AI lab of the Vrije Universiteit Brussel, the natural progression was for me to work on an even larger set of even older games for my Ph.D. research, which is described in this thesis.During this time, the department's name also changed again, to the Department of Advanced Computing Sciences (DACS).Because it seems unlikely that we will ever manage to find an even larger set of even older games than those considered in this project, I will probably not pursue an academic title beyond Ph.D. any time soon, and maybe the department's name will also remain unchanged for a longer period of time.Like most scientific endeavours, a doctoral thesis is of course the result of many people's time and efforts, and I would like to thank these people here.First of all, I wish to thank my daily supervisor, Cameron Browne, who led the Digital Ludeme Project (of which this thesis is a part), provided the opportunity for me to work on this project, and supervised and guided my research throughout the project.I also wish to thank my second supervisor, Mark Winands, who taught me several things over the years.For example, he taught me to write specific numbers in lieu of the word "several" whenever possible.Unfortunately, in this case, I did not keep count, so we will have to go with several.Of course, I would like to thank my partners in crime, Éric, Matthew, and Walter.Together, we made substantial contributions to Belgium's GDP from beer exports, in addition to contributions to AI and interdisciplinary research in the form of papers, software, talks, and workshops.Alongside Éric, I conquered the hills of Wellington, among many other adventures.I am grateful to Matthew for designing Mini Wars in Ludii, which managed to keep us entertained at night during the COVID lockdowns.Walter has always been understanding of the importance of v vi Oxford commas, naps, and correctly typesetting footnotes.1 Additional thanks go to Katrina for many delicious cookies, Luna the axolotl for being as photogenic as she is, and Okami the dog for joining Skype calls.I am very grateful to Olivier Teytaud for hosting me during my research internship with Facebook AI Research, as well as Vegard Mella who, alongside Olivier, I worked with the most.I also felt very welcomed by Jérémy Rapin.The duck on the online lunch menu sounded very appealing, but the building was unfortunately closed due to COVID.I am also grateful for the enlightening engineering and research talks by many other people.I would like to thank all the staff members of DACS for their contributions to making it a great place to work.I thank Chiara Sironi, for collaborations and discussions on game AI research.You even had to suffer through reading dozens of versions of my master's thesis.My thanks go to the regular participants of discussions in the Games research group, including Barbara Franci, Frank Thuijsman, Gijs Schoenmakers, Jos Uiterwijk, Katharina Schneider, Mathias Staudigl, and Monica Salvioli.Thanks to Rico Möckel, for being a vocal proponent of team science.Thanks also to Mayke Pothof, Iris Hoogsteder, Esther Breuls, Ellen Narinx, and many other members of the support staff, who keep things running with a lot more hard work than I am probably even consciously aware of.Special thanks to Dieudonnée van de Willige and Nava Tintarev, for liking and responding to many of my lame tweets.My thanks also go to Spyridon Samothrakis