What it is: A “fingerprinting” system that gives similar vectors similar hash codes, allowing quick similarity detection.
How it works:
- Uses special hash functions that preserve similarity relationships
- Similar vectors get placed in the same “buckets”
- Search only looks in buckets likely to contain similar items
Why it matters: Excellent for approximate search when you need speed over perfect accuracy. Great for large-scale systems.
Real-world analogy: Like sorting people by zip code - people in the same zip code are likely to be neighbors, so you only check nearby zip codes when looking for someone.