IVF Index
Also called Inverted File Index
A vector index that clusters data into buckets and only searches the few buckets near your query.
Think of it like
A warehouse split into aisles by category — you only walk the aisles that could hold what you want.
Example
FAISS with an IVF index partitions a billion vectors into cells and probes just a handful, trading a little recall for big speed.
How it actually works
IVF clusters vectors (via k-means) into cells and, at query time, searches only the nprobe nearest cells instead of everything. More probes means higher recall and slower search — a direct dial. It is frequently combined with product quantization to compress the vectors too, making it a workhorse for very large, memory-constrained collections.
For product teams
Scales search to enormous datasets on a budget; recall is a tunable you must set deliberately.
For engineers
k-means cell partitioning with nprobe-limited search; commonly IVF-PQ for compression at billion scale.
Related
- HNSW — An alternative to HNSW for ANN.
- Product Quantization — Often paired with product quantization.
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