Research map / Architectures and efficiency / RNNs and normalization
HM-ANN: Efficient Billion-Point Nearest Neighbor Search on Heterogeneous Memory
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| promote high-degree nodes to L1 instead of random nodes (ablation) | HM-ANN · billion-scale top-1 recall | query time speedup: 1.8x, 4.3x and 3.9x faster at 95%, 99%, 99.5% recall | Helped |
| add parallel L0 search only to default HNSW (ablation) | HNSW · DEEP1B top-1 | latency vs recall | No clear effect |
| add bottom-up promotion and parallel L0 search to HNSW (ablation) | HNSW on HM · DEEP1B top-1 | latency vs recall: 1.75x query time reduction at 99% recall (HM-ANN overall) | Helped |
| add dynamic migration prefetch from slow to fast memory (ablation) | HNSW + BP + PL0 · DEEP1B top-1 | latency vs recall | Helped |
| use performance-model configuration selection for efSearch (ablation) | HM-ANN · BIGANN | latency vs recall | Helped |
| explicitly manage data on heterogeneous memory (HM-ANN) vs HNSW with system-level Memory Mode / NUMA | HNSW · billion-scale top-1 recall >95% | query latency: 2x vs Memory Mode, 3.7x vs first-touch NUMA | Helped |
| HM-ANN at million-scale where data fits in DRAM | HNSW · million-scale datasets | distance computations at 99% recall: 900/query -> 850/query | Helped |
Closest papers by meaning
- LotusFilter: Fast Diverse Nearest Neighbor Search via a Learned Cutoff Table (CVPR 2025 · similarity 0.87)
- Results of the Big ANN: NeurIPS’23 competition (NeurIPS 2025 · similarity 0.86)
- Graph-Based Algorithms for Diverse Similarity Search (ICML 2025 · similarity 0.86)
- Discovering Data Structures: Nearest Neighbor Search and Beyond (NeurIPS 2025 · similarity 0.85)
- Probabilistic Routing for Graph-Based Approximate Nearest Neighbor Search (ICML 2024 · similarity 0.85)
- Distance Adaptive Beam Search for Provably Accurate Graph-Based Nearest Neighbor 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.