Given an undirected graph $G=(V,E)$ with a set of vertices $V$ and a set of edges $E$, a graph coloring problem involves finding a partition of the vertices into different independent sets. In this paper we present a new framework which combines a deep neural network with the best tools of "classical" metaheuristics for graph coloring. The proposed algorithm is evaluated on the weighted graph coloring problem and computational results show that the proposed approach allows to obtain new upper bounds for medium and large graphs. A study of the contribution of deep learning in the algorithm highlights that it is possible to learn relevant patterns useful to obtain better solutions to this problem.
翻译:鉴于一个无方向的图形$G=(V,E)$(美元),带有一套顶端值$V$和一套边缘值$E$,一个图形颜色问题涉及将顶层分割成不同的独立数据集。在本文中,我们提出了一个新框架,将深神经网络与“古典”美术学最佳图表颜色工具结合起来。在加权图形颜色问题上对提议的算法进行了评估,计算结果显示,拟议方法允许为中大图表获得新的上界值。一项有关在算法中深学习的贡献的研究突出表明,有可能学习有用的相关模式,以找到解决这一问题的更好办法。