Graph Networks Supplementary Material

The discriminator is taken to be a three-layer GCN, followed by a one-layer MLP . The non-linear activation function is Tanh and no residual connection is used. We list the information of all datasets used in the manuscript in Tab. 1. The number of nodes is fixed to 15 and the average degree of samples is 3.43. IMDB-B is a movie collaboration dataset.