Fig. 12: Dense connection process in DenseGNN. | npj Computational Materials

Fig. 12: Dense connection process in DenseGNN.

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

Fig. 12

This figure illustrates the process of densely connecting all GC layers based on the DCN strategy in DenseGNN. Each GC layer densely connects the edge-node-graph representations to those in all preceding GC layers. This design enables each GC layer to fully leverage feature information from preceding layers, facilitating effective feature reuse. By concatenating the feature maps of all layers, the shortest information propagation path is achieved. All dense connection channels are highlighted in red.

Back to article page