Simple is Better: Training an End-to-end Contract Bridge Bidding Agent without Human Knowledge

Qucheng Gong, Yu Jiang, Yuandong Tian

Keywords: Contract Bridge, Bidding, Selfplay, AlphaZero

TL;DR: State-of-the-art contract bridge bidding agent, learned from selfplay without human knowledge

Abstract: Contract bridge is a multi-player imperfect-information game where one partnership collaborate with each other to compete against the other partnership. The game consists of two phases: bidding and playing. While playing is relatively easy for modern software, bidding is challenging and requires agents to learn a communication protocol to reach the optimal contract jointly, with their own private information. The agents need to exchange information to their partners, and interfere opponents, through a sequence of actions. In this work, we train a strong agent to bid competitive bridge purely through selfplay, outperforming WBridge5, a championship-winning software. Furthermore, we show that explicitly modeling belief is not necessary in boosting the performance. To our knowledge, this is the first competitive bridge agent that is trained with no domain knowledge. It outperforms previous state-of-the-art that use human replays with 70x fewer number of parameters.