Fig. 2
From: Manifold learning and maximum likelihood estimation for hyperbolic network embedding
LaBNE+HM. An undirected, unweighted, single-component network with scale-free node degree distribution and strong clustering can be input to LaBNE+HM to reveal the hyperbolic geometry underlying it. A first hyperbolic arrangement of nodes is obtained by LaBNE (a manifold learning approach), which is later refined by HyperMap (a maximum likelihood estimation approach). In LaBNE+HM, the space of possible PS models that HyperMap has to explore is greatly reduced, as it only needs to search for appropriate node angular coordinates in a small window around the angles found by LaBNE. In the schematic, the angular coordinate of a node is refined by increasing its value