Embeddings in machine learning represent prototypes or exemplars

In machine learning, semantic embeddings of discrete items can either denote prototypes or exemplars of conceptual families. When denoting prototypes, they are informed by exemplars. In this, machine learning tacitly embraces the concept theory assumption that concept learning consists on deriving prototypes based on exemplars in context. However, when denoting exemplars, each token gets its own context-based embedding.

Backlinks