AlphaGo family algorithms are important milestones in the history of artificial intelligence. These algorithms not only solve the typical complete information game problem such as Go but also are applicable to a wider range of problems. According to their development, this paper summarizes the fundamental principle and technical characteristics for the series of algorithms from AlphaGo Fan to MuZero, elaborating how the AlphaGo family algorithms work. The key technologies employed, including the Monte Carlo tree search, modeling and training of deep neural networks, are surveyed and compared. The AlphaGo family algorithms are of significant instructive value for addressing various problems in practice, from algorithm design, neural network modeling to model utilization. This paper helps to quickly understand the principle of these algorithms and is expected to provide useful reference for the further research and development of algorithms.

artificial intelligence  /  AlphaGo family algorithms  /  Monte Carlo tree search  /  deep neural network  /  reinforcement learning