Learning to bid in bridge

摘要: Bridge bidding is considered to be one of the most difficult problems for game-playing programs. It involves four agents rather than two, including a cooperative agent. In addition, partial observability game makes it impossible predict outcome each action. this paper we present new decision-making algorithm that capable overcoming these problems. The allows models used both opponent and partners, while utilizing novel model-based Monte Carlo sampling method overcome problem hidden information. also presents learning framework uses above co-training partners. refine their selection strategies during training continuously exchange refined strategies. refinement based on inductive applied examples accumulated classes states with conflicting actions. was empirically evaluated set bridge deals. pair co-trained significantly improved performance level surpassing current state-of-the-art algorithm.