Approximate Nearest Neighbor
Also called ANN
Finding items that are almost certainly the closest, without the cost of proving it.
Think of it like
Asking a few locals for the nearest gas station instead of surveying every road on the map.
Example
A search over 50 million product embeddings returns in 10ms because it settles for "very close" rather than "provably closest".
How it actually works
Exact search means comparing your query to every stored vector — fine for thousands, hopeless for millions. ANN methods build clever indexes that skip most comparisons and return the top matches with high probability. You accept a tiny recall loss for orders-of-magnitude speedup, and you can dial the trade-off.
For product teams
It is the reason large-scale semantic features are affordable at all; the "approximate" rarely shows up in user experience.
For engineers
Sublinear-time retrieval via graph or partition indexes; recall is a tunable parameter, not a guarantee.
Related
- HNSW — HNSW is the most popular algorithm for it.
- Exact Nearest Neighbor — The honest, slower alternative that checks everything.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome