Poincaré Embeddings for Learning Hierarchical Representations (NIPS 2017)

arxiv.org

Data such as text often has a hierarchical structure. Existing embeddings in Euclidean space cannot easily model this property. Nickel & Kiela thus propose to learn embeddings in a hyperbolic space, in particular an n-dimensional Poincaré ball. The learned embeddings achieve state-of-the-art performance on determining lexical entailment and require far fewer dimensions than traditional embeddings.

Read more...
Linkedin

Want to receive more content like this in your inbox?