This document summarizes the VERSE graph embedding method. VERSE learns node embeddings by minimizing the KL divergence between the similarity distributions of a graph and its embeddings. It uses a versatile similarity notion that includes measures like personalized PageRank and SimRank. VERSE scales to large graphs with millions of nodes and edges. It produces global embeddings through noise contrastive estimation while also preserving local neighborhood information. Experiments show VERSE outperforms methods like node2vec on tasks like graph reconstruction and community detection.