Unsupervised Hyperbolic Representation Learning via Message Passing Auto-Encoders

Most of the existing literature regarding hyperbolic embedding concentrate\nupon supervised learning, whereas the use of unsupervised hyperbolic embedding\nis less well explored. In this paper, we analyze how unsupervised tasks can\nbenefit from learned representations in hyperbolic space. To explore how well\nthe hierarchical structure of unlabeled data can be represented in hyperbolic\nspaces, we design a novel hyperbolic message passing auto-encoder whose overall\nauto-encoding is performed in hyperbolic space. The proposed model conducts\nauto-encoding the networks via fully utilizing hyperbolic geometry in message\npassing. Through extensive quantitative and qualitative analyses, we validate\nthe properties and benefits of the unsupervised hyperbolic representations.\nCodes are available at https://github.com/junhocho/HGCAE.\n