Learnable Graph Convolutional Network (LGCN)

A Learnable Graph Convolutional Network (LGCN) is a Spatial Graph Convolutional Network that is a densely connected convolutional network.

• Context
  • It is composed by Learnable Graph Convolutional Layers.
• Example(s):
  • a Large-Scale Learnable Graph Convolutional Network (Gao et al., 2018),
  • …
• Counter-Example(s):
  • a Diffusion Convolutional Neural Network,
  • a PATCHY-SAN,
  • a Message Passing Neural Network,
  • a GraphSage,
  • a Spectral Graph Convolutional Network.
• See: Recurrent Neural Network, Feedforward Neural Network, Attention Mechanism.

---

---

References

2020

• (Zhou et al., 2020) ⇒ Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, and Maosong Sun (2020). "Graph neural networks: A review of methods and applications". AI Open, 1, 57-81.
  • QUOTE: The learnable graph convolutional network (LGCN) (Gao et al., 2018a) also exploits CNNs as aggregators. It performs max pooling on neighborhood matrices of nodes to get top-k feature elements and then applies 1-D CNN to compute hidden representations.

2018

• (Gao et al., 2018) ⇒ Hongyang Gao, Zhengyang Wang, and Shuiwang Ji (2018). "Large-Scale Learnable Graph Convolutional Networks". In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD 2018).
  • QUOTE: It is known that deeper networks usually yield better performance. However, prior deep models on graphs like GCNs only have two layers. While they suffer from performance loss when going deeper (...), our LGCL enables a deeper design, resulting in the learnable graph convolutional networks (LGCNs) for graph node classification. We build LGCNs based on the architecture of densely connected convolutional networks (DCNNs) (...), which achieved state-of-the-art performance in the ImageNet classification challenge (...).

Figure 3: An illustration of the proposed learnable graph convolutional network (LGCN). In this example, the nodes in the input have two features. The input feature vectors are transformed into low-dimensional representations using a graph embedding layer. After that, we stack two LGCL layers with skip concatenation connections to refine the feature vectors of each node. Finally, a fully-connected layer is used for node classification. There are three different classes in this example.

---