Research map / Graph learning / Graph algorithms and clustering
High Dimensional Similarity Search with Satellite System Graph: Efficiency, Scalability, and Unindexed Query Compatibility
Cong Fu, Changxu Wang, Deng Cai
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 |
|---|---|---|---|
| replace NSG with NSSG (angle-based pruning) | graph index · GIST1M, Crawl, GloVe at 95% precision | queries per second: 35%, 15%, and 20% speed-up on Crawl, GloVe, and GIST1M | Improved |
| replace NSG with NSSG | graph index · SIFT1M | search performance: almost the same | No clear effect |
| replace NSG with NSSG | graph index · Deep100M | queries per second: 5% to 15% speed-up over NSG | Improved |
| replace NSG with NSSG edge selection | graph index · GIST1M | edge selection time: 620+735 (s) -> 620+144 (s) | Improved |
| replace NSG with NSSG | graph index · Deep100M (single thread) | indexing time: 16 hours -> 11 hours | Improved |
| replace NSG with NSSG, parallel 16-subgraph | graph index · Deep100M | indexing time: 3.53 hours -> 2.6 hours | Improved |
| compute KNN neighbors on the fly instead of storing KNN graph | NSSG · Deep100M | memory: 37 GB less than the NSG; 56 GB at most | Improved |
| use SSG with alpha=60 instead of MRNG | graph · SIFT10K | search path length (unindexed): 4.98 -> 3.91 | Improved |
| use SSG with alpha=30 instead of alpha=60 (ablation) | SSG · SIFT10K | average out-degree: 40 -> 121 | Hurt |
| truncate SSG alpha=60 to limit degree (ablation) | SSG · SIFT10K | average out-degree: 40 -> 19 | Improved |
| use alpha=60 instead of smaller alpha (ablation) | NSSG · GIST1M, Crawl, GloVe | search performance: 60 best on all except SIFT1M where 55 is best | Improved |
| use alpha=55 instead of 60 (ablation) | NSSG · SIFT1M | search performance: 55 slightly better | No clear effect |
Closest papers by meaning
- Graph-Based Algorithms for Diverse Similarity Search (ICML 2025 · similarity 0.88)
- Distance Adaptive Beam Search for Provably Accurate Graph-Based Nearest Neighbor Search (NeurIPS 2025 · similarity 0.87)
- CSPG: Crossing Sparse Proximity Graphs for Approximate Nearest Neighbor Search (NeurIPS 2024 · similarity 0.87)
- MCGI: Manifold-Consistent Graph Indexing for Billion-Scale Disk-Resident Vector Search (NeurIPS 2026 · similarity 0.86)
- Sort Before You Prune: Improved Worst-Case Guarantees of the DiskANN Family of Graphs (ICML 2025 · similarity 0.86)
- Optimistic Query Routing in Clustering-based Approximate Maximum Inner Product Search (NeurIPS 2025 · similarity 0.85)
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.
