Trivial bundle embeddings for learning graph representations

• DOI: 10.48550/arxiv.2112.02531
• Publication Date: 2021-01-01
• Authors: Zheng Xie; Xiaojing Zuo; Yiping Song
• Keywords: Hyperbolic space; Base (topology)

Embedding real-world networks presents challenges because it is not clear how to identify their latent geometries. some disassortative networks, such as scale-free the Euclidean space has been shown incur distortions. hyperbolic spaces offer an exciting alternative but incurs distortions when embedding assortative with geometries hyperbolic. We propose inductive model that leverages both expressiveness of GCNs and trivial bundle learn node representations for or without features. A a simple case fiber bundles,a globally product its base fiber. The coordinates those can be used express factors in generating edges. Therefore, ability embeddings factors. In practice, reduces errors link prediction classification compared GCNs.