A Neural Model for Automatic Bidding of Contract Bridge
Xiaoyu Zhang, Wei Liu, Fangchun Yang · 2020
Contract bridge is a typical representative of imperfect information game, which consists of two phases: bidding and playing. One of the most challenging tasks in the contract bridge is bidding. Prior studies mainly focus on bidding without competition. However, it is not the case in reality, which inevitably leads to information loss. This work introduces hand understanding to model the real bidding process and restore the multi-task nature of the work. Therefore, Automatic Bidding Model(ABM), a novel neural architecture, is designed to solve this multi-task learning problem. A layered LSTM is provided to players to model real data, as well as an understanding and a bidding component. ABM proves its advantage through comprehensive experiments on real data set. Lastly, this study summarizes the advantage of ABM.