.png){kind=icon}
.png){kind=icon}
Return
Node-Smoothness-Based Adaptive Initial Residual Deep Graph Convolutional Network
.png){kind=icon}
PRE
AI
DOI：10.1109/JIOT.2024.3387051.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep graph convolutional networks (GCNs) can mine the deeper information of nonstructured data, e.g., capturing complex interactions within sensor topology. However, the oversmoothing problem severely limits the depth of the GCN. The initial residual can ensure the nodes retain some initial information during the propagation process, which largely alleviates the oversmoothing problem in deep GCNs. However, current works only use the grid search method to determine a fixed initial residual ratio, which cannot assign the most appropriate initial information for the nodes with different over-smoothnesses. This article proposes a novel method named node-smoothness-based adaptive initial residual deep GCN (NSAIR-GCN). Specifically, it can be divided into two processes: 1) considering the oversmoothing from another perspective, i.e., determining whether nodes are oversmoothed based on the difference in node representations before and after updating and identifying those severely oversmoothed nodes and 2) assigning appropriate initial residual ratios to these nodes based on their smoothness. Extensive semi-supervised node classification experiments on several standard data sets have shown that the adaptive initial residual ratio determined by node smoothness performs better than the previous fixed initial residual ratio and achieves the state of the art.
Keywords:
Convolutional neural networks
Measurement
Internet of Things
Smoothing methods
Feature extraction
Vectors
Symmetric matrices
Deep graph convolutional network (GCN)
initial residual
oversmoothing
oversmoothing metrics
Journal
IF:
8.9
Papers:
1.4W
Citations:
7.8W