Master the Game of 3D Connect Four

Category:  AI
Features
  • MCTS
  • Self-play learning and Neural Network
  • Evaluation Function Design
  • Prior + UCB
  • Dominant Pruning

In this project, we introduce a method dominant pruning. This idea combines the concept "induction" in game theory. Assume the opponent is rational, while running MCTS, if we can confirm a move dominate the game, then, by induction, the opponent won't pick up its parent move. Therefore, we can prune its parent node to improve the efficiency of MCTS.

Programming Languages/Frameworks
  • C++
  • Python
Team Members
Go to GitHub for research details