.png){kind=icon}
.png){kind=icon}
Return
Information-controlled graph convolutional network for multi-view semi-supervised classification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neunet.2024.107102.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Graph convolutional networks have achieved remarkable success in the field of multi-view learning. Unfortunately, most graph convolutional network-based multi-view learning methods fail to capture long-range dependencies due to the over-smoothing problem. Many studies have attempted to mitigate this issue by decoupling graph convolution operations. However, these decoupled architectures lead to the absence of feature transformation module, thus limiting the expressive power of the model. To this end, we propose an information-controlled graph convolutional network for multi-view semi-supervised classification. In the proposed method, we maintain the paradigm of node embeddings during propagation by imposing orthogonality constraints on the feature transformation module. By further introducing a damping factor based on residual connections, we theoretically demonstrate that the proposed method can alleviate the over-smoothing problem while retaining the feature transformation module. Furthermore, we prove that the proposed model can stabilize both forward inference and backward propagation in graph convolutional networks. Extensive experimental results on benchmark datasets demonstrate the effectiveness of the proposed method.
Keywords:
Multi-view learning
Semi-supervised classification
Graph convolutional network
Layer normalization
Journal
IF:
6.3
Papers:
7.8K
Citations:
3.0W