inter-agent attention实际上是在计算agent i对于各个group中在可视范围内的agent j的关注程度,并将其aggregate起来作为对group k的representation c. Inter-group Attention (Hierarchical State Representation Using Multi-Graph Attention) Inter-agent Attention得到了agent i对每个cluster的information aggregation,则在grou...
Recently, there has been a promising tendency to generalize convolutional neural networks (CNNs) to graph domain. However, most of the methods cannot obtain adequate global information due to their shallow structures. In this paper, we address this challenge by proposing a hierarchical graph ...
4.4.3 Segment-level Update 最后,我们使用 Graph Attention Network [19] (GAT) 对段节点之间的关系进行建模,如下所示 其中AS 是分段节点的二进制邻接矩阵。 4.5 Learning and Discussion 在我们的模型中,各种节点嵌入(NS、NR、NZ)、分配矩阵(AS R、ARZ)和涉及的组件参数是模型参数。请注意,每个 GAT 或 GCN ...
Recently, there has been a promising tendency to generalize convolutional neural networks (CNNs) to graph domain. However, most of the methods cannot obtain adequate global information due to their shallow structures. In this paper, we address this challenge by proposing a hierarchical graph attenti...
此文提出了一个名为Hierarchical Graph Convolution Networks的结构用于交通预测,此结构有node-level的micro-layer捕捉节点间的关系,也有region-level的macro-layer捕捉区域间的关系。区域是由节点聚类产生的。 此文创建了node-level和region-level间的信息交互来同时获取两个级别的信息而不是简单的fusion。
Graph Neural Networks (GNNs) draw their strength from explicitly modeling the topological information of structured data. However, existing GNNs suffer from limited capability in capturing the hierarchical graph representation which plays an important role in graph classification. In this paper, we innovat...
A Hierarchical Graph Attention Network for Stock Movement Prediction. As we conducted experiments on two different tasks, node classification and graph classification, we provide two different version of codes for each tasks. Please refer to our paperHATS: A Hierarchical Graph Attention Network for Sto...
Document Modeling with Graph Attention Networks for Multi-grained Machine Reading Comprehension 论文解读:使用图注意力网络进行多粒度机器阅读理解的文档建模 阅读目的:学习该论文中基于文档结构的多粒度建模。 摘要 “自然问题”是一种具有挑战性的新机器阅读理解基准,它具有两个粒度的答案,即长答案(通常是一个段落...
Combinatorial Optimization by Graph Pointer Networks and Hierarchical Reinforcement Learning URL https://arxiv.org/pdf/1911.04936.pdf TL;DR 设计了一种图指针网络分层强化学习模型,用于解决带约束的组合优化类问题(例如带时间窗的TSP问题)。作者在低层和高层人为设定了强化学习的目标,利用低层生成可行解,高层以...
Skeleton-based action recognition has drawn much attention recently. Previous methods mainly focus on using RNNs or CNNs to process skeletons. But they ignore the topological structure of the skeleton which is very important for action recognition. Recently, Graph Convolutional Networks (GCNs) achieve...