首页> 外文会议>International Joint Conference on Web Intelligence and Intelligent Agent Technology >GGTAN: Graph Gated Talking-Heads Attention Networks for Traveling Salesman Problem
【24h】

GGTAN: Graph Gated Talking-Heads Attention Networks for Traveling Salesman Problem

机译:GGTAN:Graph Gated Talking-Headings旅行推销人员问题的关注网络

获取原文

摘要

Traveling Salesman Problem (TSP) is one of the most typical NP-hard combinatorial optimization problems with a variety of real-life applications. In this paper, we propose a Graph Gated Talking-Heads Attention Networks (GGTAN) trained with reinforcement learning (RL) for tackling TSP. GGTAN can learn characteristic structure information better by introducing talking-heads attention mechanism and a gated convolutional sub-network, which make hidden information moving across between attention heads and control each attention head's importance respectively, unlike recently proposed models which use attention mechanism for solving TSP. Experimental results on TSP up to 100 nodes demonstrate that our model obtains shorter tour lengths than other learning-based methods under the same solve strategy for problem instances of fixed graph sizes, and achieves better generalization on variable graph sizes compared with recent state-of-the-art models on the optimality gap.
机译:旅行推销员问题(TSP)是具有各种现实生活应用的最典型的NP硬组合优化问题之一。 在本文中,我们提出了一个图形通话,带有加固学习(RL)的GARGET-HEARMS注意网络(GGTAN),用于解决TSP。 GGTAN通过引入通知注意机制和门控卷积子网络可以更好地学习特征结构信息,该子网络使隐藏的信息分别在关注头部之间移动并控制每个关注头的重要性,与最近提出的模型,它使用注意机制来解决TSP 。 TSP高达100个节点的实验结果表明,我们的模型比其他基于学习的方法获得更短的巡视长度,而在相同的解决方法的解决策略下,与最近的最新状态相比,在变量图大小上实现了更好的泛化 最优性差距上的最新模型。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号