Variational Inference for Training Graph Neural Networks in Low-Data Regime through Joint Structure-Label Estimation
Danning Lao *, Xinyu Yang *, Qitian Wu, Junchi Yan
Shanghai Jiao Tong University
What Is Graph?
Graph is an important representation of natural structures.
Graph Neural Network
Challenge: graph structure that can be used in aggregation is often incomplete
Challenge: node labels that can be used as�supervision can be very limited
GCN and GAT’s performance degrades notably with incomplete structure and limited labels
Can we design a weakly-supervised graph neural network jointly estimating graph structure and node label?
Formulate our problem setting
Variational inference for weakly-supervised graph learning
Two-branch architecture for label-structure joint inference
Results on Cora
Results on Citeseer
Results on Pubmed
Results under extreme low-data regime
Case study on Infectious Disease Spreading
Ablation study on model architecture
Node classification visualization
Link prediction visualization
Thanks for your attention!