There were three competitions held as part of IJCAI-PRICAI2020 in January this year. One of these was the Mahjong Competition in which competitors were tasked with developing an intelligent Mahjong agent that was able to compete with other agents, as well as human players, on an online AI platform.
Mahjong is a four-player tile-based imperfect information game that originated in China. It is a game of skill, strategy, and calculation, and it also involves a degree of chance.
This introduction to the competition by Yushan Zhou includes an explanation of how to play Mahjong.
Following the preliminary round, 16 teams made it through to the final stage of the competition. All of these teams prepared video presentations describing their models and methodology. You can watch these on YouTube at this playlist.
The winner of the competition was team SuperJong. The team used a reinforcement learning algorithm and their model architecture comprised a convolutional neural network. You can find out more about this from the team’s video below:
You can watch replays of the games at this page. Just click on the hyperlinks in the “name” column to access the various rounds. Then, click on the “matches” tab to see the video links for each match.
The competition was jointly organised by AI Lab, Institute of Network Computing and Information Systems, Peking University, and WEIZHIYU (Beijing) Technology Co., Ltd.
The competition webpage can be found here.