M4: Extensive-Form Computation
When games have imperfect information—when you cannot see an opponent's move—computing equilibria becomes substantially harder. Information sets, sequential rationality, and subgame perfect equilibria formalize reasoning under partial observation. This is the computational home of poker, and also the setting where LLM agents must decide with incomplete information about other agents' states and intentions. PA3 applies CFR to Kuhn poker, an algorithm central to AI game-playing.
Module preview
When games have imperfect information—when you cannot see an opponent's move—computing equilibria becomes substantially harder. Information sets, sequential rationality, and subgame perfect equilibria formalize reasoning under partial observation. This is the computational home of poker, and also the setting where LLM agents must decide with incomplete information about other agents' states and intentions. PA3 applies CFR to Kuhn poker, an algorithm central to AI game-playing.
Lectures and materials
L9: Absent-Minded Driver and Information Sets
Imperfect recall and decisions under partial observation.
L10: Extensive-Form Nash and Subgame Perfect Equilibria
Sequential rationality and credible strategies.
Programming assignment
PA3: CFR and Kuhn Poker · Due November 9 at 11:59 PM CDT