ScaNN is an open-source library for efficient vector similarity search, outperforming other libraries in speed and accuracy for large datasets.
ScaNN (Scalable Nearest Neighbors) is a vector similarity search library developed by Google Research. It addresses the computational challenge of quickly finding nearest dataset embeddings for a given query embedding, especially when datasets are too large for exhaustive search and high dimensionality makes pruning difficult. The library utilizes a novel compression technique called anisotropic vector quantization, which significantly boosts accuracy compared to prior works by optimizing for nearest-neighbor search accuracy rather than just minimizing average distance between vectors and their quantized forms. This technique is particularly effective for Maximum Inner-Product Search (MIPS), a common method for defining query-database embedding similarity. ScaNN's anisotropic vector quantization penalizes quantization error that is parallel to the original vector more heavily, as this type of error disproportionately impacts high inner products, which are critical for accurate MIPS estimation. This allows ScaNN to better estimate inner products likely to be in the top-k MIPS results, leading to higher accuracy and faster query processing.
Integrations: TensorFlow, Numpy
Platforms: Web
View full ScaNN profile on Tools-Radar | Browse Coding & Developer Tools tools | Alternatives to ScaNN
Tools-Radar is a free directory of 10,000+ AI tools — discover, compare, and choose the right AI software for your needs. Visit tools-radar.com