Matryoshka Embeddings

Definitions

1. Embeddings designed so truncated lower-dimensional prefixes still retain useful semantic information, making one vector usable at multiple dimensionalities.

   In plain English: Embedding vectors built so shorter versions still work reasonably well.

   Example: "Matryoshka embeddings let us trade retrieval quality for speed without retraining."

Related Terms

• Vector Database
• Vector Database
• Semantic Search
• Approximate Nearest Neighbor
• Contrastive Learning
• Dense Retrieval
• Dual Encoder
• Knowledge Graph Embedding