ALT The high-level overview of SmartGD. Component (a) sketches the training procedure of the GAN-based model. The self-challenging mechanism is explained in component (c), which is applied only when the optimization criterion is given. Component (b) describes the inference procedure for drawing unseen graphs.

ALT he qualitative evaluation of 7 SmartGD models by comparing with 5 competitive and representative benchmarks. All the graphs presented above are unseen during the training phase of SmartGD. The name of the graphs with the number of nodes N and the number of edges M is presented in the row header. The colors of the nodes represent their community within the graph, computed by Girvan-Newman algorithm [37]. For each layout, the computation time t (without including the pre-processing time) is reported in seconds. All the computation time is computed on the CPU except the SmartGD for Rome graphs and DeepGD for Rome graphs.