Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Hierarchy Aware Features for Reducing Mistake Severity
Ashima Garg, Depanshu Sani, Saket Anand
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 |
|---|---|---|---|
| add geometric consistency loss L_gc to soft-consistency+margin (ablation) | HAF · CIFAR-100 | mistake severity: 2.24 -> 2.22 | Improved |
| add margin loss L_m to soft consistency + geometric consistency (ablation) | HAF · CIFAR-100 | mistake severity: 2.32 -> 2.22 | Improved |
| add margin loss L_m to soft consistency + geometric consistency (ablation) | HAF · CIFAR-100 | hierarchical distance@5: 1.66 -> 1.40 | Improved |
| add soft hierarchical consistency L_shc to cross-entropy (ablation) | HAF · CIFAR-100 | hierarchical distance@5: 2.24 -> 1.61 | Improved |
| add soft hierarchical consistency L_shc to cross-entropy (ablation) | HAF · CIFAR-100 | top-1 error (%): 22.11 -> 22.70 | Hurt |
| add L_gc to soft consistency + margin loss (ablation) | HAF · CIFAR-100 | top-1 error (%): 22.12 -> 22.25 | Hurt |
| add full HAF losses to cross-entropy (ablation) | HAF · iNaturalist-19 | mistake severity: 2.39 -> 2.27 | Improved |
| add full HAF losses to cross-entropy (ablation) | HAF · iNaturalist-19 | top-1 error (%): 36.48 -> 36.47 | No clear effect |
| add L_gc to soft consistency alone (ablation) | HAF · iNaturalist-19 | top-1 error (%): 36.23 -> 36.60 | Hurt |
| train with HAF instead of cross-entropy | Cross-Entropy · CIFAR-100 | mistake severity: 2.35 -> 2.24 | Improved |
| train with HAF instead of cross-entropy | Cross-Entropy · tieredImageNet-H | mistake severity: 7.05 -> 7.03 | No clear effect |
| apply CRM test-time post-processing to HAF | HAF · CIFAR-100 | mistake severity: 2.24 -> 2.23 | No clear effect |
Closest papers by meaning
- Hier-COS: Making Deep Features Hierarchy-aware via Composition of Orthogonal Subspaces (CVPR 2026 · similarity 0.87)
- Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 2024 · similarity 0.86)
- Hierarchical Multi-Label Classification Networks (ICML 2018 · similarity 0.86)
- Self Iterative Label Refinement via Robust Unlabeled Learning (NeurIPS 2025 · similarity 0.83)
- ULAREF: A Unified Label Refinement Framework for Learning with Inaccurate Supervision (ICML 2024 · similarity 0.83)
- Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions (CVPR 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.
