Engineering2026-09-172 min read

contrastive-dual-encoders

VDaily Team
Maintainer

Contrastive & dual encoders

📂 model-architectures

MODEL ARCHITECTURES#

Contrastive & dual encoders

Encode two inputs independently (such as a query and document or image and caption) and train matching pairs to land near each other.

MENTAL MODEL#

Two readers meet in a shared coordinate system; they are fast because candidates do not interact until their vectors are compared.

DATA FLOW#

  • Paired inputs A and B

  • Separate or shared encoders

  • Projected normalized vectors

  • Similarity matrix

  • Contrastive matching loss

How it trains#

Batch contrastive objectives reward paired examples and treat other examples as negatives. Negative quality, duplicate semantics, and temperature materially affect the learned space.

How inference runs#

Encode each side independently and compare vectors with dot product or cosine similarity. Precompute the large candidate side; optionally rerank top results with a cross-encoder.

Strengths#

  • • Scales retrieval to large corpora

  • • Aligns modalities without a joint decoder

  • • Enables zero-shot classification through label text in some settings

Trade-offs#

  • • Independent encoding misses fine-grained cross-input interactions

  • • False negatives can distort training

  • • Global similarity may ignore spatial, temporal, or compositional details

Use it when#

  • Fast retrieval or matching is the first stage

  • One side can be indexed offline

  • Recall is followed by task-appropriate reranking when needed

Avoid or challenge it when#

  • Every candidate needs deep token-to-token comparison

  • The task depends on precise spatial relationships

  • Raw similarity scores are assumed to be probabilities

Illustrative published families#

  • • CLIP image–text dual encoder

  • • Dense passage retrieval

  • • Bi-encoder semantic search

Commonly combines with#

PRIMARY REFERENCES#

Model Architectures#

Model Architectures#

Tags:
contrastive-dual-encoders — Blog — VDaily