Research map / Label-efficient and robust learning / Recognition and long-tail
ELIAS: End-to-End Learning to Index and Search in Large Output Spaces
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 |
|---|---|---|---|
| Learn cluster-to-label assignments end-to-end instead of keeping them fixed (ablation) | ELIAS · Amazon-670K | R@100: 61.72 -> 68.95 | Helped |
| Learn cluster-to-label assignments end-to-end instead of keeping them fixed (ablation) | ELIAS · Amazon-670K | P@1: 46.63 -> 48.68 | Helped |
| Add a sparse ranker without score calibration (ablation) | ELIAS Stage 2 · Amazon-670K | P@1: 48.68 -> 50.72 | Helped |
| Add score calibration to the sparse ranker (ablation) | ELIAS with sparse ranker · Amazon-670K | P@1: 50.72 -> 51.41 | Helped |
| Use an ensemble of 3 models (ablation) | ELIAS with score calibration · Amazon-670K | P@1: 51.41 -> 53.02 | Helped |
| Set shortlist-loss weight λ to 0 instead of 0.05 (ablation) | ELIAS-1 · Amazon-670K | R@100: 68.95 -> 66.05 | Hurt |
| Increase shortlist-loss weight from 0 to 0.05 (ablation) | ELIAS-1 · Amazon-670K | R@100: 66.05 -> 68.95 | Helped |
| Reduce adjacency row sparsity κ from 1000 to 100 (ablation) | ELIAS-1 · Amazon-670K | R@100: 68.95 -> 56.38 | Hurt |
| Prune learned adjacency edges with threshold 0.25 (ablation) | ELIAS-1 · Amazon-670K | P@1: 48.68 -> 48.65 | No clear effect |
| Retain only the top 500 label assignments per cluster (ablation) | ELIAS-1 · Amazon-670K | R@100: 68.95 -> 68.95 | No clear effect |
Closest papers by meaning
- Extreme Multi-label Classification from Aggregated Labels (ICML 2020 · similarity 0.89)
- A no-regret generalization of hierarchical softmax to extreme multi-label classification (NeurIPS 2018 · similarity 0.88)
- Label Disentanglement in Partition-based Extreme Multilabel Classification (NeurIPS 2021 · similarity 0.87)
- ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces (ICML 2025 · similarity 0.86)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.85)
- OAK: Enriching Document Representations using Auxiliary Knowledge for Extreme Classification (ICML 2024 · 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.