Research map / Label-efficient and robust learning / Recognition and long-tail
Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View
Jinping Wang, Zixin Tong, Zhiwu Xie, Zhiqiang Gao
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 batch-wise inverse reweighting (ablation) | CE · CIFAR-100-LT IF=100 | Top-1 accuracy: 41.6 -> 46.7 | Improved |
| add macro-level reweighting (ablation) | CE · CIFAR-100-LT IF=100 | Top-1 accuracy: 41.6 -> 45.4 | Improved |
| add macro-level reweighting on top of batch-wise inverse (ablation) | CE · CIFAR-100-LT IF=100 | Top-1 accuracy: 46.7 -> 47.9 | Improved |
| replace cross-entropy by inverse-view reweighting | ResNet-32 · CIFAR-100-LT IF=100 | Top-1 accuracy: 41.62 -> 47.88 | Improved |
| replace inverse-frequency weighting by inverse-view reweighting | ResNet-32 · CIFAR-100-LT IF=200 | Top-1 accuracy: 25.25 -> 43.14 | Improved |
| add inverse-view loss reweighting | GLMC · CIFAR-100-LT IF=100 | Top-1 accuracy: 55.9 -> 58.3 | Improved |
| add inverse-view loss reweighting | CE* · CIFAR-10-LT IF=200 | Top-1 accuracy: 69.1 -> 78.6 | Improved |
| add learning-rate schedule on top of loss reweighting | CE*+Loss · CIFAR-10-LT IF=100 | Top-1 accuracy: 82.3 -> 82.0 | No clear effect |
| add learning-rate schedule on top of loss reweighting | CE*+Loss · CIFAR-100-LT IF=100 | Top-1 accuracy: 47.9 -> 48.9 | Improved |
| add learning-rate schedule on top of loss reweighting | CE*+Loss · iNaturalist 2018 | Top-1 accuracy: 66.3 -> 65.8 | Hurt |
| add inverse-view loss reweighting | CE* · ImageNet-LT Few | Top-1 accuracy: 4.6 -> 8.3 | Improved |
| add inverse-view loss reweighting | GLMC · ImageNet-LT All | Top-1 accuracy: 56.3 -> 57.6 | Improved |
Closest papers by meaning
- Neural Collapse for Cross-entropy Class-Imbalanced Learning with Unconstrained ReLU Features Model (ICML 2024 · similarity 0.89)
- Neural Collapse To Multiple Centers For Imbalanced Data (NeurIPS 2024 · similarity 0.87)
- Inverse Weight-Balancing for Deep Long-Tailed Learning (AAAI 2024 · similarity 0.87)
- Distribution Alignment Optimization through Neural Collapse for Long-tailed Classification (ICML 2024 · similarity 0.86)
- Guiding Neural Collapse: Optimising Towards the Nearest Simplex Equiangular Tight Frame (NeurIPS 2024 · similarity 0.86)
- Thumb on the Scale: Optimal Loss Weighting in Last Layer Retraining (NeurIPS 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.
