人工智能培训

搜索

深度学习论文:双曲图注意力网络(Hyperbolic Graph Attention Network)

[复制链接]
sulee 发表于 2019-12-9 13:17:49 | 显示全部楼层 |阅读模式
sulee 2019-12-9 13:17:49 196 0 显示全部楼层
深度学习论文:双曲图注意力网络(Hyperbolic Graph Attention Network)图神经网络(GNN)在处理图形方面表现出卓越的性能,最近引起了相当大的研究关注。但是,大多数现有的GNN模型主要是为欧几里得空间中的图设计的。最近的研究已经证明,图形数据具有非欧几里得潜伏的解剖结构。不幸的是,到目前为止,在非欧几里得环境中很少有关于GNN的研究。为了弥合这一差距,本文首次尝试在双曲空间中研究GNN注意机制。双曲GNN的研究面临一些独特的挑战:由于双曲空间不是向量空间,因此无法进行向量运算(例如向量加法,减法和标量乘法)。为了解决这个问题,我们使用陀螺矢量空间,它为双曲几何提供了优雅的代数形式,以变换图形中的特征。此外,由于双曲空间中的数学运算比欧几里德空间中的数学运算更复杂,因此,我们进一步设计了一种使用对数和指数映射的新型加速策略,以提高所提出模型的效率。通过与其他最新基准方法进行比较,在四个实际数据集上的综合实验结果证明了我们提出的双曲线图关注网络模型的性能。
Graph neural network (GNN) has shown superior performance in dealing withgraphs, which has attracted considerable research attention recently.However,most of the existing GNN models are primarily designed for graphs in Euclideanspaces.Recent research has proven that the graph data exhibits non-Euclideanlatent anatomy.Unfortunately, there was rarely study of GNN in non-Euclideansettings so far.To bridge this gap, in this paper, we study the GNN withattention mechanism in hyperbolic spaces at the first attempt.The research ofhyperbolic GNN has some unique challenges: since the hyperbolic spaces are notvector spaces, the vector operations (e.g., vector addition, subtraction, andscalar multiplication) cannot be carried.To tackle this problem, we employ thegyrovector spaces, which provide an elegant algebraic formalism for hyperbolicgeometry, to transform the features in a graph;and then we propose thehyperbolic proximity based attention mechanism to aggregate the features.Moreover, as mathematical operations in hyperbolic spaces could be morecomplicated than those in Euclidean spaces, we further devise a novelacceleration strategy using logarithmic and exponential mappings to improve theefficiency of our proposed model.The comprehensive experimental results onfour real-world datasets demonstrate the performance of our proposed hyperbolicgraph attention network model, by comparisons with other state-of-the-artbaseline methods.深度学习论文:双曲图注意力网络(Hyperbolic Graph Attention Network)
URL地址:https://arxiv.org/abs/1912.03046     ----pdf下载地址:https://arxiv.org/pdf/1912.03046    ----深度学习论文:双曲图注意力网络(Hyperbolic Graph Attention Network)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

sulee当前离线
新手上路

查看:196 | 回复:0

快速回复 返回顶部 返回列表