Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss
Kaidi Cao, Colin Wei, Adrien Gaidon, Nikos Arechiga, Tengyu Ma
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 |
|---|---|---|---|
| Use LDAM-DRW instead of RW | Imbalanced IMDB review classification · IMDB validation | mean error: 23.66 -> 17.84 | Improved |
| Add deferred re-weighting to LDAM (ablation) | iNaturalist 2018 · iNaturalist 2018 validation | Top-1 error: 35.42 -> 32.00 | Improved |
| Add deferred re-weighting to ERM (ablation) | iNaturalist 2018 · iNaturalist 2018 validation | Top-1 error: 42.86 -> 36.27 | Improved |
| Use LDAM instead of ERM cross-entropy (ablation) | Imbalanced CIFAR-10, long-tailed, ratio 100 · Imbalanced CIFAR-10, long-tailed, ratio 100 | Top-1 validation error: 29.64 -> 26.65 | Improved |
| Add DRW to LDAM (ablation) | Imbalanced CIFAR-10, long-tailed, ratio 100 · Imbalanced CIFAR-10, long-tailed, ratio 100 | Top-1 validation error: 26.65 -> 22.97 | Improved |
| Replace class-independent margins with label-distribution-aware margins (ablation) | CIFAR-100 with DRW, long-tailed, ratio 100 · Imbalanced CIFAR-100, long-tailed, ratio 100 | Top-1 validation error: 59.49 -> 57.96 | Improved |
| Use DRW with LDAM rather than LDAM with the standard SGD schedule (ablation) | Imbalanced Tiny ImageNet, long-tailed, ratio 100 · Imbalanced Tiny ImageNet, long-tailed, ratio 100 | Top-1 validation error: 64.04 -> 62.53 | Improved |
| Use deferred re-sampling instead of deferred re-weighting (ablation) | Imbalanced CIFAR-10, long-tailed, ratio 100 · Imbalanced CIFAR-10, long-tailed, ratio 100 | Top-1 validation error: 25.14 -> 25.50 | No clear effect |
| Use deferred re-weighting rather than standard ERM training (ablation) | Imbalanced Tiny ImageNet, step, ratio 100 · Imbalanced Tiny ImageNet, step, ratio 100 | Top-1 validation error: 63.82 -> 62.36 | Improved |
| Use LDAM rather than ERM cross-entropy (ablation) | Imbalanced Tiny ImageNet, step, ratio 10 · Imbalanced Tiny ImageNet, step, ratio 10 | Top-1 validation error: 50.89 -> 49.08 | Improved |
| Apply Class-Balanced Softmax rather than ERM (ablation) | Imbalanced Tiny ImageNet, step, ratio 100 · Imbalanced Tiny ImageNet, step, ratio 100 | Top-1 validation error: 63.82 -> 74.90 | Hurt |
| Replace LDAM with uniform-margin cross-entropy in the DRW method (ablation) | Imbalanced CIFAR-10, step, ratio 100 · Imbalanced CIFAR-10, step, ratio 100 | Top-1 validation error: 23.08 -> 27.67 | Hurt |
Closest papers by meaning
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.87)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.87)
- Inverse Weight-Balancing for Deep Long-Tailed Learning (AAAI 2024 · similarity 0.87)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.86)
- Learning with Noisy Labels Using Hyperspherical Margin Weighting (AAAI 2024 · similarity 0.86)
- Revive Re-weighting in Imbalanced Learning by Density Ratio Estimation (NeurIPS 2024 · 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.
