Symmetric Spaces for Graph Embeddings

Abstract:

Learning faithful graph representations as sets of vertex embeddings has become a fundamental intermediary step in a wide range of machine learning applications. We propose the systematic use of symmetric spaces in representation learning, a class encompassing many of the previously used embedding targets. This enables us to introduce new methods for graph embeddings with an enhanced representation capacity. We develop a tool to analyze the embeddings and infer structural properties of the data sets. Our approach outperforms competitive baselines for graph reconstruction tasks on various synthetic and real-world datasets, and on two downstream tasks.