.png){kind=icon}
返回
A new framework for graph neural network with local information diffusion
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-021-03132-5.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Graph neural networks have attracted a lot of attention in recent years. In most graph neural networks, the representation of a node is obtained by aggregating its information and the information of its neighbor nodes. However, a common disadvantage for many existing graph neural networks is that the depth of the network is limited, because the node representation often appears excessively smooth after only a few layers. In this paper, a new graph neural network based on local information diffusion is proposed, in which the representations of nodes are extracted by a cascaded architecture of network layers. In each layer, first, the proposed model randomly samples some nodes as signal sources and the representations of the signal sources are extracted by a convolutional neural network. The representations of other nodes are then updated by the local information diffusion from signal sources rather than traditional information aggregation. To further reduce the number of parameters, a parameter sharing mechanism is introduced, i.e. some network layers are grouped into a block and the parameters of neural network layers in a block are the same. In addition, a strategy is proposed to automatically estimate the hyper-parameters of the network. The experiments on some benchmark datasets show that the depth of our model is bigger than the traditional models and the performance is also better than the traditional models. It indicates that our model is effective and competitive.
Keyword:
Local information diffusion
Parameter sharing
Hyperparameter estimation
Deeper structure
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
引用论文
Rh(II)-catalyzed intramolecular annulation of N-sulfonyl 1,2,3-triazoles with indole derivatives: a new method for synthesis pyranoindoles.png)
Tetrahedron
TetrahedronIF0
PREAI
分享收藏Refined measurement of SecA-driven protein transport reveals indirect coupling to ATP turnover
OAAI
IF0
OAAI分享收藏分享收藏
查看更多内容