Research on algorithmic approaches to solving incomplete/imperfect information games, with focus on CFR family algorithms and their application to poker.

Documents

DocumentDescription
MCCFR & Nash EquilibriumCore algorithms for solving imperfect information games
Kuhn Poker BenchmarkSimplified poker game used as CFR testing benchmark
Implementation GuideStep-by-step tutorial to build a CFR poker AI

Key Concepts

  • Incomplete Information: Players don’t know all aspects of the game (e.g., opponent types, payoffs)
  • Imperfect Information: Players can’t observe all moves made (e.g., hidden cards in poker)
  • Nash Equilibrium: Strategy profile where no player can improve by unilaterally changing their strategy
  • Counterfactual Regret: Regret calculated as if trying to reach a game state, isolating decision quality
  • Regret Matching: Choose actions proportional to accumulated positive regret

Applications

  • Poker AI (Libratus, Pluribus)
  • Economic modeling
  • Security games
  • Negotiation systems
  • Any multi-agent decision problem with hidden information