Fig. 8: A demonstration of the node aggregation and updating in GNN.
From: A geometric-information-enhanced crystal graph network for predicting properties of materials
a The full picture of one step of forward propagation in GNN. b The detailed process in two nodes.\(\,{v}_{i}^{t}\) means the ith node in the tth layer,\(\,{u}_{{ij}}^{t}\) means the edge between node \(i\) and \({j}\) in the tth layer (in this work, \({u}_{{ij}}^{t}\) doesn’t vary with \(t\)), W is a learnable parameter matrix, \(f\) and \(g\) are nonlinear functions.
