NSG

NSG, short for Navigating Spreading-out Graph, is a graph-based ANN algorithm introduced in a 2018 paper by Fu et al. that achieves competitive recall and latency with HNSW while consuming substantially less memory. The algorithm builds a single-layer graph (unlike HNSW's hierarchical layers) using a monotonic search property that guarantees greedy traversal will not get stuck in local minima. NSG and its successor SSG (Satellite System Graph) have been benchmarked competitively with HNSW on standard datasets, sometimes with lower memory footprint for the same recall. The algorithm is implemented in some research-oriented vector libraries and has been adopted in production at Alibaba and other Chinese tech companies. NSG remains less widely deployed in commercial vector databases than HNSW, partly because HNSW's earlier publication and Facebook's adoption gave it network-effect dominance. AI governance teams evaluating alternative graph-based indexes consider NSG when memory cost is the dominant constraint and they have the engineering resources to validate the alternative implementation.

Alternative indexes with Centralpoint: Centralpoint stays model-agnostic across whatever ANN algorithm your vector backend supports — HNSW, NSG, NGT, DiskANN — letting you evaluate and switch as workload economics evolve. Tokens are metered per skill, prompts stay local, and the chatbot fleet embeds across portals with one line of JavaScript.


Related Keywords:
NSG,,