Research map / Label-efficient and robust learning / Recognition and long-tail
Making Better Mistakes: Leveraging Class Hierarchies With Deep Networks
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 |
|---|---|---|---|
| Apply HXE with α=0.5 instead of flat cross-entropy | ResNet-18 · tieredImageNet-H | hierarchical distance of mistake: 6.97±0.005 -> 6.58±0.036 | Helped |
| Apply HXE with α=0.5 instead of flat cross-entropy | ResNet-18 · tieredImageNet-H | top-1 error: 31.58±0.137 -> 37.24±0.490 | Hurt |
| Apply soft labels with β=5 instead of flat cross-entropy | ResNet-18 · tieredImageNet-H | average hierarchical distance @5: 5.63±0.004 -> 5.19±0.007 | Helped |
| Apply soft labels with β=5 instead of flat cross-entropy | ResNet-18 · tieredImageNet-H | top-1 error: 31.58±0.137 -> 37.29±0.064 | Hurt |
| Apply HXE with α=0.5 instead of flat cross-entropy | ResNet-18 · iNaturalist-19 | hierarchical distance of mistake: 2.43±0.003 -> 2.16±0.001 | Helped |
| Apply HXE with α=0.5 instead of flat cross-entropy | ResNet-18 · iNaturalist-19 | top-1 error: 44.04±0.197 -> 52.06±0.225 | Hurt |
| Apply soft labels with β=10 instead of flat cross-entropy | ResNet-18 · iNaturalist-19 | average hierarchical distance @20: 2.87±0.007 -> 1.99±0.005 | Helped |
| Apply soft labels with β=10 instead of flat cross-entropy | ResNet-18 · iNaturalist-19 | top-1 error: 44.04±0.197 -> 55.54±0.173 | Hurt |
| Use the real class hierarchy rather than a randomized hierarchy | HXE and soft labels · iNaturalist-19 | hierarchical distance | Helped |
| Output class probabilities and derive conditionals in the loss rather than outputting conditional probabilities directly (ablation) | HXE · Evaluation datasets | performance across metrics | Helped |
Closest papers by meaning
- When does dough become a bagel? Analyzing the remaining mistakes on ImageNet (NeurIPS 2022 · similarity 0.89)
- Automated Classification of Model Errors on ImageNet (NeurIPS 2023 · similarity 0.88)
- No Cost Likelihood Manipulation at Test Time for Making Better Mistakes in Deep Networks (ICLR 2021 · similarity 0.88)
- Understanding the detrimental class-level effects of data augmentation (NeurIPS 2023 · similarity 0.87)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.86)
- Few-Class Arena: A Benchmark for Efficient Selection of Vision Models and Dataset Difficulty Measurement (ICLR 2025 · similarity 0.85)
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.