Analyzing simplified Geister using DREAM
Lucien Troillet, Kiminori Matsuzaki · 2021 IEEE Conference on Games (CoG) · 2021
Geister is a board imperfect information game created in Germany and presenting an interesting challenge for the field of artificial intelligence. In this study we apply DREAM (Deep Regret Minimization with Advantage Baselines and Model-free Learning), a neural-network variation of Counterfactual Regret Minimization developed by Steinberger et al., to multiple variants of Geister. This paper shows a methodological approach of evaluating game strategies on different variants of Geister and illustrates the possible generalizability of the DREAM algorithm on other board games.