Figure 7:

Performance in community detection of a Gaussian mixture clustering applied to a common set of vertex latent positions estimated with different methods, for a sample of m = 40 two-block multilayer SBM graphs, with n = 256 vertices, and four classes of connectivity matrices. The class separation controls the magnitude of the smallest eigenvalue of the graphs, and as this increases, all methods show better accuracy. MRDPG, which is based on a non-convex optimization problem, performance the best for small α, but as long as α is large enough, MASE and OMNI also show a great performance.