Advanced Deep Learning with Python

Convolutional Graph Networks

Convolutional Graph Networks (ConvGNN) use a stack of special graph convolutional layers (Gconv*) to perform a convolution over the nodes of a graph when updating the state vectors. In a similar way to GraphNNs, the graph convolution takes the neighbors of a node and produces its vector representation . But whereas GraphNN uses the same layer (that is, the same set of weights) over all steps t of the computation of , ConvGNN uses different layers at every step. The difference between the two approaches is illustrated ...