Definition
An information retrieval architecture where queries and documents are processed independently into vector representations within a shared latent space, enabling high-speed similarity search at the cost of capturing fine-grained interactions between query and document tokens.
Unlike Cross-Encoders, Dual-Encoders allow document embeddings to be pre-computed and indexed for sub-millisecond retrieval.
"Two separate parallel scanners creating unique barcodes for a question and a book so they can be instantly matched without re-reading the book."
Conceptual Overview
An information retrieval architecture where queries and documents are processed independently into vector representations within a shared latent space, enabling high-speed similarity search at the cost of capturing fine-grained interactions between query and document tokens.
Disambiguation
Unlike Cross-Encoders, Dual-Encoders allow document embeddings to be pre-computed and indexed for sub-millisecond retrieval.
Visual Analog
Two separate parallel scanners creating unique barcodes for a question and a book so they can be instantly matched without re-reading the book.