GNN解决什么问题：
针对non-euclidean数据
限制：
浅层语义
结构的拓展
特征融合
非固定点探索

Graph Types

Directed Graphs
methods: use two kinds of weight matrix, to incorporate more precise structural information.
Heterogeneous Graphs

  1. The simplest way to process heterogeneous graph is to convert the type of each node to a one-hot feature vector which is concatenated with the original feature.
  2. Introduces the concept of metapath into the propagation on the heterogeneous graph. With metapath, we can group the neighbors according to their node types and distances. For each neighbor group, GraphInception treats it as a sub-graph in a homogeneous graph to do propagation and concatenates the propagation results from different homogeneous graphs to do a collective node representation.
  3. Heterogeneous attention network

Graphs with Edge Information

  1. we can convert the graph to a bipartite graph where the original edges also become nodes and one original edge is split into two new edges which means there are two new edges between the edge node and begin/end nodes.
  2. we can adapt different weight matrices for the propagation on different kinds of edges.

Dynamic Graphs
First collect spatial infor- mation by GNNs, then feed the outputs into a sequence model like sequence-to-sequence model or CNNs. Differ- ently, Structural-RNN [42] and ST-GCN

Propagation Types

Convolution
Gate
Attention

Training methods

Sampling
Receptive Field Control
Data Augmentation

  1. Co-Training
  2. Self-Training

Unsupervised Training

  1. Graph AutoEncoder
  2. Graph GAN

General Framework

  1. Message Passing Neural Network
  2. Non Local Neural Network
  3. Graph Networks