Graph Coloring problem solving using monte carlo tree search and deep reinforcement learning

Wenzhu Yang, Zhanshan Li

Abstract: The graph coloring problem, as a well-known NP-hard problem, holds significant value in practical applications. In this paper, a self-learning method that combines Monte Carlo tree search with deep reinforcement learning is proposed to efficiently solve the graph coloring problem. This method offers two principal advantages. Firstly, it leverages deep reinforcement learning to eliminate the necessity for manual feature construction and data labeling. Secondly, by combining the neural network with Monte Carlo tree search, the neural network can provide comprehensive guidance based on the structural information of the graph, facilitating a more effective balance between exploration and exploitation, thereby leading to superior solutions. Finally, experimental results demonstrate that the method proposed herein has distinct advantages over existing graph coloring algorithms. Moreover, this approach also exhibits outstanding performance when dealing with graph instances whose vertex size surpasses those encountered during the training phase.

External IDs:dblp:journals/jco/YangL25