Deep learning approaches to the game of Connect6

Abstract

Especially with the rise of artificial intelligence, people need more ability to think logically and make judgments than before. Board games can be used as a tool for training logic, which has a positive effect on the sustainable development of society. Connect6 is a very popular board game because it is a fair and highly complex game with simple rules. This research aims to address the technical challenges of building a game-based logical learning system with a deep neural network for Connect6. We used heuristic knowledge to establish the knowledge base. Our novel algorithm, based on residual deep convolutional neural networks (DCNNs), outperforms conventional approaches. We compare neural network architectures, demonstrating the superiority of residual DCNNs.Our AI program, “Kavalan,” achieves remarkable search performance. Furthermore, the introduced approaches lend themselves to application in other sudden-death board games, such as Gomoku, the endgame of Chess and Shogi.