Order Embedding
A class of representation learning methods that map partial order relations into a vector space. The embedding of an ancestor concept is required to be no greater than that of its descendant in every dimension, which converts hierarchical structure into vector containment and can be optimized with a hinge loss.
Timeline (last 90 days)
PEEL is a lightweight probing method that operates on frozen embeddings, decoding a classification DAG from a concept vocabulary via an order-embedding objective and a minimum spanning arborescence algorithm.
Order embeddings map the ancestor relationship between concepts to a vector containment relationship: if concept A is an ancestor of concept B, then A's embedding vector is no greater than B's embedding vector in every dimension.
All Facts (2)
PEEL is a lightweight probing method that operates on frozen embeddings, decoding a classification DAG from a concept vocabulary via an order-embedding objective and a minimum spanning arborescence algorithm.
50%UnverifiedOrder embeddings map the ancestor relationship between concepts to a vector containment relationship: if concept A is an ancestor of concept B, then A's embedding vector is no greater than B's embedding vector in every dimension.
50%