Graph Structure Learning in GNN

We extract the definitions in Wu et al., 2020¹. Given node embeddings $\mathbf E_i$¹,

$$ \begin{align} \mathbf M_i &= \tanh(\alpha \mathbf E_i \Theta_i) \\ \mathbf A &= \operatorname{ReLU}(\tanh(\alpha (\mathbf M_1 \mathbf M_2^T - \mathbf M_2\mathbf M_1^T))), \end{align} $$

The author also proposed sparse requirement and only take the top-$k$ largest elements in $A$.

Wu2020 Wu Z, Pan S, Long G, Jiang J, Chang X, Zhang C. Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks. arXiv [cs.LG]. 2020. Available: http://arxiv.org/abs/2005.11650 ↩︎ ↩︎

Planted: 2022-11-28 by L Ma;

References:

Wu2020 Wu Z, Pan S, Long G, Jiang J, Chang X, Zhang C. Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks. arXiv [cs.LG]. 2020. Available: http://arxiv.org/abs/2005.11650

Dynamic Backlinks to cards/forecasting/gnn-graph-structure-learning:

Graph Structure Learning in GNN

We can learn a graph structure without prior knowledge

L Ma (2022). 'Graph Structure Learning in GNN', Datumorphism, 11 April. Available at: https://datumorphism.leima.is/cards/forecasting/gnn-graph-structure-learning/.