Research map / Label-efficient and robust learning / Recognition and long-tail
Hyperbolic Space with Hierarchical Margin Boosts Fine-Grained Learning from Coarse Labels
ShuLin Xu, Yifan Sun, Faen Zhang, Anqi Xu, Xiu-Shen Wei, Yi Yang
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 |
|---|---|---|---|
| replace Euclidean embedding with hyperbolic embedding (ablation) | coarse-label supervised baseline · LIVING-17 all-way | accuracy: 32.71 -> 33.78 | Improved |
| replace Euclidean embedding with hyperbolic embedding (ablation) | coarse-label supervised baseline · CIFAR-100 all-way | accuracy: 24.92 -> 26.19 | Improved |
| add Hierarchical Cosine Margin (HCM) loss in hyperbolic space (ablation) | hyperbolic baseline · LIVING-17 all-way | accuracy: 33.78 -> 46.53 | Improved |
| add Hierarchical Cosine Margin (HCM) loss in hyperbolic space (ablation) | hyperbolic baseline · CIFAR-100 all-way | accuracy: 26.19 -> 33.22 | Improved |
| add Adaptive Hierarchical Cosine Distance (AHCD) strategy (ablation) | HCM · LIVING-17 all-way | accuracy: 46.53 -> 53.09 | Improved |
| add Adaptive Hierarchical Cosine Distance (AHCD) strategy (ablation) | HCM · CIFAR-100 all-way | accuracy: 33.22 -> 36.28 | Improved |
| add AHCD on top of HCM (ablation) | HCM · CIFAR-100 5-way | accuracy: 77.64 -> 81.42 | Improved |
| PE-HCM vs SCGM-A | PE-HCM · LIVING-17 all-way | accuracy: 49.31 -> 53.09 | Improved |
| PE-HCM vs ANCOR | PE-HCM · ENTITY-13 all-way | accuracy: 42.29 -> 41.64 | Hurt |
| PE-HCM vs fine upper-bound | PE-HCM · CIFAR-100 all-way | accuracy: 31.35 -> 36.28 | Improved |
| PE-HCM vs MaskCon re-implementation, Recall@1 | PE-HCM · CIFAR-100 retrieval | Recall@1: 65.35 -> 62.11 | Hurt |
| PE-HCM vs MaskCon re-implementation, mAP | PE-HCM · CIFAR-100 retrieval | mAP: 60.27 -> 62.04 | Improved |
Closest papers by meaning
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · similarity 0.88)
- Learning Fine-grained Domain Generalization via Hyperbolic State Space Hallucination (AAAI 2025 · similarity 0.87)
- Unsupervised Feature Learning with Emergent Data-Driven Prototypicality (CVPR 2024 · similarity 0.86)
- Understanding Fine-tuning CLIP for Open-vocabulary Semantic Segmentation in Hyperbolic Space (CVPR 2025 · similarity 0.86)
- Hyperbolic Category Discovery (CVPR 2025 · similarity 0.86)
- Compositional Entailment Learning for Hyperbolic Vision-Language Models (ICLR 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.
