Search has played a crucial role in recent successes in games like Chess, Go and Poker. However, search methods for these games rely heavily on being able to explicitly enumerate states in the player's information state, which becomes intractable in games like Stratego or Hearthstone. Currently, state-of-the-art agents in these games avoid this limitation by foregoing search altogether. Instead, they learn a policy directly using methods such as policy gradients. The thesis aims to investigate search methods for this type of games.
Seznam odborné literatury
Sutton, Richard S. and Andrew G. Barto. “Reinforcement Learning: An Introduction.” IEEE Trans. Neural Networks 9 (1998): 1054-1054.
Nisan, Noam et al. “Algorithmic Game Theory.” (2007).