Research map / Label-efficient and robust learning / Self-supervised and contrastive
Use All the Labels: A Hierarchical Multi-Label Contrastive Learning Framework
Check 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 hierarchical multi-label HiMulConE rather than the strongest reported baseline | ResNet-50 representation learning · ImageNet | top-1 accuracy: 78.70 -> 79.14 | Helped |
| Use hierarchical multi-label HiMulConE rather than the strongest reported baseline | ResNet-50 representation learning · DeepFashion | top-1 accuracy: 72.82 -> 73.21 | Helped |
| Use hierarchical multi-label HiMulConE rather than the strongest reported baseline | ResNet-50 representation learning · iNaturalist | top-1 accuracy: 57.33 -> 59.40 | Helped |
| Use hierarchical multi-label HiMulConE rather than the strongest reported baseline | ResNet-50 representation learning · ModelNet40 | top-1 accuracy: 83.49 -> 88.46 | Helped |
| Use HiMulConE rather than Guided | ResNet-50 representation learning · DeepFashion unseen | top-1 accuracy: 74.04 -> 75.29 | Helped |
| Use HiMulConE rather than Guided | ResNet-50 representation learning · ModelNet40 unseen | top-1 accuracy: 82.22 -> 85.37 | Helped |
| Combine level-weighted HiMulCon with hierarchy enforcement as HiMulConE rather than SupCon | ResNet-50 retrieval embeddings · DeepFashion | MAP: 31.5 -> 35.6 | Helped |
| Combine level-weighted HiMulCon with hierarchy enforcement as HiMulConE rather than SupCon | ResNet-50 retrieval embeddings · iNaturalist | MAP: 61.5 -> 66.9 | Helped |
| Combine level-weighted HiMulCon with hierarchy enforcement as HiMulConE rather than SupCon | ResNet-50 retrieval embeddings · ModelNet40 | MAP: 21.6 -> 26.0 | Helped |
| Use HiConE instead of combined HiMulConE (ablation) | ResNet-50 representation learning · DeepFashion unseen | top-1 accuracy: 75.28 vs 75.29 | No clear effect |
| Use HiMulCon instead of combined HiMulConE (ablation) | ResNet-50 representation learning · DeepFashion unseen | top-1 accuracy: 75.29 -> 74.88 | Hurt |
| Use HiMulCon instead of combined HiMulConE (ablation) | ResNet-50 representation learning · ModelNet40 unseen | top-1 accuracy: 85.37 -> 84.44 | Hurt |
Closest papers by meaning
- Multi-Label Supervised Contrastive Learning (AAAI 2024 · similarity 0.90)
- Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 2024 · similarity 0.88)
- Weakly-Supervised Contrastive Learning for Imprecise Class Labels (ICML 2025 · similarity 0.88)
- Self-Supervised Contrastive Learning is Approximately Supervised Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- Multi-Label Learning with Stronger Consistency Guarantees (NeurIPS 2024 · similarity 0.86)
- How does Labeling Error Impact Contrastive Learning? A Perspective from Data Dimensionality Reduction (ICML 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.