Research map / Learning theory and algorithms / Prediction and decision losses
Learning to Hash Robustly, Guaranteed
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 |
|---|---|---|---|
| Replace uniform coordinate sampling with min-max optimized hash distribution at each node (exponent rho=0.83) | LSH forest on MNIST subset · MNIST subset (750 pts) | minimum success probability: 0.35 -> 0.63 (about 1.8x) | Helped |
| Replace uniform coordinate sampling with min-max optimized hash distribution at each node (rho=1) | LSH forest on ImageNet subset · ImageNet subset (624 pts) | bottom 10% success probability: 0.275 -> 0.576 (about 2.1x) | Helped |
| Replace uniform sampling with optimized hashing (rho=1) | LSH forest on ImageNet subset · ImageNet subset (624 pts) | average success probability: 0.621 -> 0.772 | Helped |
| Replace uniform sampling with optimized hashing (rho=0.83) on MNIST | LSH forest on MNIST subset · MNIST subset (750 pts) | average success probability: 0.737 -> 0.878 | Helped |
| Exponent rho varied across optimized runs (rho=1, 0.25, 0.1 vs 0.83) (ablation) | optimized hashing on MNIST · MNIST subset (750 pts) | minimum success probability: rho=0.83: 0.63; rho=1: 0.6; rho=0.25: 0.42; rho=0.1: 0.36 (uniform 0.35) | Mixed |
| Use optimized hashing only at a subset of nodes (e.g. final levels) instead of all nodes | optimized hashing · datasets around 10^5 points | query-time improvement | No clear effect |
| Optimize hash distribution on a single-cluster dataset with independent coordinates | LSH forest · theoretical single-cluster model | worst-case success probability | No clear effect |
Closest papers by meaning
- Falconn++: A Locality-sensitive Filtering Approach for Approximate Nearest Neighbor Search (NeurIPS 2022 · similarity 0.88)
- Optimal Densification for Fast and Accurate Minwise Hashing (ICML 2017 · similarity 0.88)
- Local Density Estimation in High Dimensions (ICML 2018 · similarity 0.88)
- Approximate Nearest Neighbor Search through Modern Error-Correcting Codes (ICLR 2023 · similarity 0.88)
- KALAHash: Knowledge-Anchored Low-Resource Adaptation for Deep Hashing (AAAI 2025 · similarity 0.87)
- Discovering Data Structures: Nearest Neighbor Search and Beyond (NeurIPS 2025 · 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.