A Graph Neural Network Assisted Monte Carlo Tree Search Approach to Traveling Salesman Problem

Zhihao Xing, Shikui Tu

Keywords: Traveling Salesman Problem, Graph Neural Network, Monte Carlo Tree Search

TL;DR: A Graph Neural Network Assisted Monte Carlo Tree Search Approach to Traveling Salesman Problem

Abstract: We present a graph neural network assisted Monte Carlo Tree Search approach for the classical traveling salesman problem (TSP). We adopt a greedy algorithm framework to construct the optimal solution to TSP by adding the nodes successively. A graph neural network (GNN) is trained to capture the local and global graph structure and give the prior probability of selecting each vertex every step. The prior probability provides a heuristics for MCTS, and the MCTS output is an improved probability for selecting the successive vertex, as it is the feedback information by fusing the prior with the scouting procedure. Experimental results on TSP up to 100 nodes demonstrate that the proposed method obtains shorter tours than other learning-based methods.