Research map / Label-efficient and robust learning / Recognition and long-tail
Distributional Robustness Loss for Long-Tail Learning
Dvir Samuel, Gal Chechik
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 a robustness margin with a shared radius (eps>0) (ablation) | DRO-LT · CIFAR-100-LT (imbalance 100) | top-1 accuracy: 43.92 -> 45.66 | Improved |
| scale the radius by 1/sqrt(class size) instead of sharing it (ablation) | DRO-LT · CIFAR-100-LT (imbalance 100) | top-1 accuracy: 45.66 -> 46.92 | Improved |
| learn a per-class radius instead of eps/sqrt(n) (ablation) | DRO-LT · CIFAR-100-LT (imbalance 100) | top-1 accuracy: 46.92 -> 47.31 | Improved |
| full learned radius vs no margin (ablation) | DRO-LT · CIFAR-100-LT (imbalance 50) | top-1 accuracy: 52.31 -> 57.57 | Improved |
| full learned radius vs no margin (ablation) | DRO-LT · CIFAR-100-LT (imbalance 10) | top-1 accuracy: 59.54 -> 63.41 | Improved |
| full learned radius vs no margin (ablation) | DRO-LT · ImageNet-LT | top-1 accuracy: 48.1 -> 53.5 | Improved |
| replace CE training with DRO-LT (learned eps) | ResNet-32 · CIFAR-100-LT (imbalance 100), Few classes | few-shot accuracy: 7.4 -> 23.8 | Improved |
| replace CE training with DRO-LT (learned eps) | ResNet-32 · CIFAR-100-LT (imbalance 100), Many classes | many-shot accuracy: 65.5 -> 64.7 | No clear effect |
| use DRO-LT representation instead of CE representation, with cRT classifier (ablation) | ResNet-32 · CIFAR-100-LT (imbalance 100) | top-1 accuracy: 41.2 -> 47.3 | Improved |
| use DRO-LT representation instead of CE representation, with CE classifier (ablation) | ResNet-32 · CIFAR-100-LT | top-1 accuracy: 38.5 -> 41.2 | Improved |
| use DRO-LT representation instead of CE representation, with LWS classifier (ablation) | ResNet-32 · CIFAR-100-LT | top-1 accuracy: 41.4 -> 46.8 | Improved |
| learn representation with balanced resampling instead of CE (ablation) | ResNet-32 · CIFAR-100-LT (CE classifier) | top-1 accuracy: 38.5 -> 34.9 | Hurt |
Closest papers by meaning
- Long-Tailed Recognition via Information-Preservable Two-Stage Learning (NeurIPS 2025 · similarity 0.88)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.88)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.88)
- TAET: Two-Stage Adversarial Equalization Training on Long-Tailed Distributions (CVPR 2025 · similarity 0.87)
- Long-tailed Adversarial Training with Self-Distillation (ICLR 2025 · similarity 0.87)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.87)
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.
