siamese-network-shared-weights-paired-embedding
IN premise — summaries/2026/08/24/wiki-Latent_space.md
Created 2026-08-24T17:11:18+00:00
A Siamese network architecture uses two identical sub-networks with shared weights to process paired inputs, training a loss function that enforces closeness for similar pairs and separation for dissimilar pairs in embedding space
Summary
This describes how a Siamese network learns to judge similarity: it feeds two items through the same set of shared parameters and trains by nudging the outputs of matching items closer together while pushing non-matching items apart. The practical upshot is that the system ends up with a learned distance measure, so it can reliably tell "these two are the same kind of thing" versus "these two are not," which is the foundation for any matching, retrieval, or comparison task the system performs.