Research map / Graph learning / Graph algorithms and clustering
Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs
Yu. A. Malkov, Dmitry A. Yashunin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use exponential layer assignment with mL=1/ln(M) instead of mL=0 (single layer) (ablation) | Hierarchical NSW · 10M random d=4 vectors | query time at fixed recall | Improved |
| add hierarchy (mL>0) on very high dimensional data (ablation) | Hierarchical NSW · 100k random d=1024 vectors | query time at fixed recall | Improved |
| use neighbor selection heuristic instead of closest neighbors (ablation) | Hierarchical NSW · clustered and low-dimensional data | query time at fixed recall | Improved |
| use neighbor selection heuristic instead of closest neighbors (ablation) | Hierarchical NSW · uniform very high dimensional data | query time at fixed recall | No clear effect |
| set Mmax0 to 2*M instead of M (ablation) | Hierarchical NSW · 5M SIFT learn | query time at fixed recall | Improved |
| raise Mmax0 above 2*M (ablation) | Hierarchical NSW · 5M SIFT learn | query time at fixed recall | Hurt |
| use larger M (ablation) | Hierarchical NSW · high recall / high dimensional data | query time at fixed recall | Mixed |
| raise efConstruction beyond 100 (ablation) | Hierarchical NSW · 10M SIFT | index quality vs build time | No clear effect |
| add hierarchical layers to NSW (ablation) | NSW baseline · d=4 and d=8 random hypercube | query time at fixed recall | Improved |
| add hierarchical layers to NSW (ablation) | NSW baseline · Wiki-8 JS-divergence (general space) | distance computations: almost 3 orders of magnitude | Improved |
Closest papers by meaning
- Graph-Based Algorithms for Diverse Similarity Search (ICML 2025 · similarity 0.90)
- Distance Adaptive Beam Search for Provably Accurate Graph-Based Nearest Neighbor Search (NeurIPS 2025 · similarity 0.88)
- Navigable Graphs for High-Dimensional Nearest Neighbor Search: Constructions and Limits (NeurIPS 2024 · similarity 0.88)
- Sort Before You Prune: Improved Worst-Case Guarantees of the DiskANN Family of Graphs (ICML 2025 · similarity 0.86)
- CSPG: Crossing Sparse Proximity Graphs for Approximate Nearest Neighbor Search (NeurIPS 2024 · similarity 0.86)
- MCGI: Manifold-Consistent Graph Indexing for Billion-Scale Disk-Resident Vector Search (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
