VColRL: Learn to solve the Vertex Coloring Problem using Reinforcement Learning

TMLR Paper5083 Authors

Abstract: We propose VColRL, a deep reinforcement learning framework for solving the Vertex Coloring Problem (VCP), which aims to color the vertices of a graph using the minimum number of colors such that no two adjacent vertices share the same color. VColRL is based on a novel Markov Decision Process (MDP) formulation, identified through a systematic evaluation of multiple configurations. It employs a reduction-based neural architecture and a reward mechanism designed to minimize the highest-numbered color used from an ordered set. Experiments on synthetic and benchmark graphs show that VColRL consistently outperforms greedy and learning-based methods in terms of color usage, while achieving competitive performance with advanced optimization solvers and search-based baselines. In addition to delivering high-quality solutions, VColRL achieves significantly faster runtimes than the baselines, demonstrating strong scalability and generalization across diverse graphs.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Christopher_Mutschler1

Submission Number: 5083