Hyperbolic embedding
Embedding (often Graph embedding) that uses Hyperbolic space rather than Euclidean space.
Hyperbolic space is nice for embedding networks because it’s a perfect space for embedding a branching tree (more space as we move farther away from the center) and it also naturally incorporate community structure (angle direction).
See also Krioukov2010hyperbolic as an intro.
Methods
- Poincaré Embedding
- “ultrahyperbolic” embedding - Law2020ultrahyperbolic
- Lorentzian: Law2019Lorentzian
- Chami2019hyperbolic - hyperbolic Graph convolutional network
- Chami2020low dimensional - Embedding of Knowledge graph