Fig. 1
From: GTAT: empowering graph neural networks with cross attention
GTAT framework. Given a graph \(\mathcal {G}\) with \(N\) nodes, along with a set of node feature representations H, we first obtain the GDV of these nodes through the TFE. Subsequently, we use MLP transforms GDV into a set of topology representations T. GTAT layer receives \(\mathcal {G}\) and these two representations as input, then transforms and outputs two updated sets of representations. Finally, based on the set of node feature representations, our model outputs the predictions of nodes’ classifications.
