Artificial Intelligence

Embedding

Definition

A list of numbers representing a word, image or molecule, arranged so that similar things sit close together.

Everything a model processes has to become numbers first. What makes an embedding useful rather than arbitrary is the geometry: distance in the space corresponds to similarity in whatever the model was trained to care about. That is also its limitation — similar for one task is not similar for another.

Where this comes up

Articles that define embedding in context.

Hi! Need help? Chat with NSTC ✨