Distributed Network Embedding: A Distributed Control Approach for Undirected Networks

Abstract

Listen

Translate article

In this paper, a new network embedding method is proposed for an effective network representation of vertices. While the existing embedding methods have embedded a set of vertices via network representation learning in a fully centralized way, we present a novel distributed network embedding (DNE) method without representation learning. By coupling first-order proximity and second-order proximity with distance embedding constraints, a distributed and explicit embedding protocol steers a set of vertices in a vector space to preserve local and global structural information of networks. The proposed method does not require representation learning process, and is distributed, explicit, and straightforward. Therefore, we expect our approach to be fast, cost-effective, especially suitable for dynamic networks, and robust for sparse networks. Lastly, numerical experiments will briefly validate the proposed embedding method using a well-known Zachary's karate club network.