Research map / Architectures and efficiency / RNNs and normalization
Extreme Classification in Log Memory using Count-Min Sketch: A Case Study of Amazon Search with 50M Products
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 |
|---|---|---|---|
| use MACH with B=20K, R=32 rather than Parabel with 16 trees | Amazon Search Dataset · Amazon Search Dataset, wRecall@100 | weighted recall: 0.5810 -> 0.6541 | Helped |
| use MACH with B=20K, R=32 rather than DSSM, 256 dim | Amazon Search Dataset · Amazon Search Dataset, wRecall@100 | weighted recall: 0.441 -> 0.6541 | Helped |
| use MACH, B=20K, R=32 rather than the embedding baseline | Amazon Search Dataset matching · Amazon Search Dataset, matching | recall_weighted: 0.8391 -> 0.8486 | No clear effect |
| use MACH rather than Parabel | Amazon Search Dataset ranking · Amazon Search Dataset, ranking | prec@1_weighted: 0.8788 -> 0.9109 | Helped |
| use MACH, B=32, R=25 rather than OAA | ODP · ODP | accuracy: 9% -> 15.446% | Helped |
| use MACH, B=512, R=20 rather than OAA | Fine-grained ImageNet · Fine-grained ImageNet | accuracy: 17% -> 10.675% | Hurt |
| use MACH, B=2000 rather than DisMEC | Wiki10-31K · Wiki10-31K | P@1: 0.8520 -> 0.8544 | Helped |
| reduce MACH bucket count from B=20K to B=10K (ablation) | Amazon Search Dataset · Amazon Search Dataset, wRecall@100 | weighted recall: 0.6541 -> 0.6419 | Hurt |
| replace the unbiased estimator with the min estimator (ablation) | MACH · ODP | accuracy: 15.446 -> 12.212 | Hurt |
| replace the unbiased estimator with the median estimator (ablation) | MACH · ODP | accuracy: 15.446 -> 14.434 | Hurt |
| replace the unbiased estimator with the min estimator (ablation) | MACH · Fine-grained ImageNet | accuracy: 10.675 -> 9.743 | Hurt |
| replace the unbiased estimator with the median estimator (ablation) | MACH · Fine-grained ImageNet | accuracy: 10.675 -> 10.713 | No clear effect |
Closest papers by meaning
- HashAttention: Semantic Sparsity for Faster Inference (ICML 2025 · similarity 0.84)
- ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces (ICML 2025 · similarity 0.84)
- Benchmarking and Building Long-Context Retrieval Models with LoCo and M2-BERT (ICML 2024 · similarity 0.83)
- On the Recall Scaling Laws in Mamba: A Theoretical and Mechanistic Study via Hashing (NeurIPS 2026 · similarity 0.83)
- Memory Type Varies: Empowering LLM Agents for Long-Term Memory with Diverse Strategies (NeurIPS 2026 · similarity 0.83)
- NUDGE: Lightweight Non-Parametric Fine-Tuning of Embeddings for Retrieval (ICLR 2025 · similarity 0.83)
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.