Locality-Sensitive Hashing
A family of hashing techniques designed so similar objects collide with higher probability than dissimilar ones, allowing approximate-neighbor candidates to be found without scanning the full dataset.
Conventional hashes try to spread even slightly different inputs across unrelated outputs. Locality-sensitive hashing deliberately does the opposite with respect to a chosen similarity measure: nearby objects are more likely to land in the same or related buckets, shrinking the candidate set for search.
LSH is a framework rather than one universal algorithm. Jaccard, cosine, and Euclidean spaces use different constructions, and recall/latency depends on table count, hash length, and probing policy. FAISS index architecture for vector similarity search provides a useful modern ANN comparison point.