Keywords: spatial state-action features, general game playing, reinforcement learning, feature discovery

"Learning state-action features for general game playing"

One of the core challenges considered in the field of artificial intelligence research is the development of algorithms and computer programs that can autonomously plan, make decisions, and achieve goals. Games are often used as test domains for such research, as they are typically thought to require a certain level of intelligence to play well. Aside from this use of games as test domains, the development of techniques that can automatically learn to play games can also contribute to anthropological studies of games as cultural artefacts, as well as game design and development industries. The research in this thesis led to the development of game-independent algorithms that can autonomously discover relevant patterns for a wide variety of different board games. These patterns can be used to improve the level of play of computer programs, and they can also provide humans with insights into any tactics or strategies they encode.              

Language: English