Figure 6:

Average normalized mean squared error of estimating the expected adjacency matrix of a sample of graphs using different embedding methods. Graphs in the sample are distributed as a multilayer SBM with n = 256 vertices, two communities, four different classes of connectivity matrices, and a training and test samples with m = 40 graphs. MASE and JE are the only methods that are flexible enough to capture the heterogeneous structure of the graphs, but MASE shows superior performance among these two. As the graphs become more different, the error of MRDPG and OMNI increases.