Graph Learning Network (GLN)

GLN creates a set of node embeddings H(l) that are later combined to produce an intermediary representation H_int​(l). Then, we use the updated node information with the adjacency information to produce a local embedding of the nodes' information H_local(l)​ that is also the output H(l+1). We also broadcast the information of the local embedding to produce a global embedding H_global​(l). We combine the local and global embeddings to predict the next layer adjacency A(l+1). Additionally, we create three Synthetic Graph Datasets: the 3D-Surface, Community, and Geometric Figures. The source code is available in the public repository https://gitlab.com/mipl/graph-learning-network and the datasets are available in https://gitlab.com/mipl/graph-learning-network/-/tree/master/datasets.