Research map / Learning theory and algorithms / Prediction and decision losses
Learning to Index for Nearest Neighbor Search
Chih‐Yi Chiu, Amorntip Prayoonwong, Yin-Chih Liao
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 |
|---|---|---|---|
| rank clusters by NN probability instead of query-centroid distance (ablation) | first-level retrieval · SIFT1B and DEEP1B | top-1 recall | Improved |
| use query-centroid similarity feature (Prob_QCS) rather than raw query data (ablation) | probability-based ranking · SIFT1B and DEEP1B | top-1 recall | Improved |
| concatenate raw query data with query-centroid similarities (ablation) | Prob_QCS · SIFT1B and DEEP1B | top-1 recall | No clear effect |
| add cluster quantity estimation proportional to NN probability (ablation) | Prob_RAW · SIFT1B and DEEP1B | top-1 recall | Improved |
| add hierarchical learning for second-level clusters (ablation) | Prob_RAW with normalization · SIFT1B and DEEP1B | top-1 recall | Improved |
| compare sum, standard and exponent normalization of quantity estimates (ablation) | HProb_RAW · SIFT1B and DEEP1B | top-1 recall | No clear effect |
| index PCA-compressed 32d data instead of original 128d (ablation) | HProb_RAW+gamma_std · SIFT1B RVQ 4096x4096 | top-1 recall (1 cluster): 0.779 -> 0.803 | Improved |
| index PCA-compressed 32d data instead of original 96d (ablation) | HProb_RAW+gamma_std · DEEP1B RVQ 4096x4096 | top-1 recall (1 cluster): 0.753 -> 0.756 | Improved |
| increase first-level codebook size from 256 to 16384 (ablation) | Prob_RAW · SIFT1B RVQ | top-1 recall (15 clusters): 0.637 -> 0.871 | Improved |
Closest papers by meaning
- SOAR: Improved Indexing for Approximate Nearest Neighbor Search (NeurIPS 2023 · similarity 0.87)
- LoRANN: Low-Rank Matrix Factorization for Approximate Nearest Neighbor Search (NeurIPS 2024 · similarity 0.85)
- Graph-Based Algorithms for Diverse Similarity Search (ICML 2025 · similarity 0.85)
- Probabilistic Routing for Graph-Based Approximate Nearest Neighbor Search (ICML 2024 · similarity 0.85)
- Discovering Data Structures: Nearest Neighbor Search and Beyond (NeurIPS 2025 · similarity 0.85)
- Optimistic Query Routing in Clustering-based Approximate Maximum Inner Product Search (NeurIPS 2025 · similarity 0.84)
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.
