Fig. 3: Input graph and attribute update steps in DenseGNN. | npj Computational Materials

Fig. 3: Input graph and attribute update steps in DenseGNN.

From: DenseGNN: universal and scalable deeper graph neural networks for high-performance property prediction in crystals and molecules

Fig. 3

This figure illustrates the input graph for DenseGNN and the process of updating node, edge, and graph attributes. The input graph consists of node attributes, edge attributes, and optional graph attributes. During the first update step, edge attributes are updated based on information from the nodes forming the edges, the graph attributes, and the previous edge attributes. In the subsequent steps, node attributes are updated incorporating information from the edge attributes and graph attributes, while graph attributes are updated considering information from both node and edge attributes. This iterative update process optimizes the representation of the graph structure and its associated features.

Back to article page