Research map / Label-efficient and robust learning / Recognition and long-tail
Long-tail learning via logit adjustment
Read the paperCheck 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 |
|---|---|---|---|
| Add class-prior logit offsets during training | ResNet-32 · CIFAR-10-LT | balanced error: 27.16 -> 22.33 | Helped |
| Add class-prior logit offsets during training | ResNet-32 · CIFAR-100-LT | balanced error: 61.64 -> 56.11 | Helped |
| Add class-prior logit offsets during training | ResNet-50 · ImageNet-LT | balanced error: 53.11 -> 48.89 | Helped |
| Add class-prior logit offsets during training | ResNet-50 · iNaturalist | balanced error: 38.66 -> 33.64 | Helped |
| Tune the post-hoc class-prior adjustment parameter τ (ablation) | ResNet-32 · CIFAR-10-LT | balanced error: 22.60 -> 19.08 | Helped |
| Tune the post-hoc class-prior adjustment parameter τ (ablation) | ResNet-32 · CIFAR-100-LT | balanced error: 58.24 -> 57.90 | Helped |
| Tune the post-hoc class-prior adjustment parameter τ (ablation) | ResNet-50 · ImageNet-LT | balanced error: 49.66 -> 49.56 | Helped |
| Tune the post-hoc class-prior adjustment parameter τ (ablation) | ResNet-50 · iNaturalist | balanced error: 33.98 -> 33.80 | Helped |
| Combine logit-adjustment margin with the adaptive margin (ablation) | ResNet-50 · iNaturalist | balanced error: 33.64 -> 31.56 | Helped |
| Combine logit-adjustment margin with the adaptive margin (ablation) | ResNet-152 · iNaturalist | balanced error: 31.15 -> 29.22 | Helped |
| Train the combined logit-adjustment and adaptive-margin loss for 200 epochs instead of 90 (ablation) | ResNet-152 · iNaturalist | balanced error: 29.22 -> 28.02 | Helped |
| Apply deferred reweighting to the adaptive-margin method (ablation) | ResNet-50 · iNaturalist | balanced error: 33.6 -> 32.0 | Helped |
Closest papers by meaning
- Robust Logit Adjustment for Learning with Long-Tailed Noisy Data (AAAI 2025 · similarity 0.93)
- Multiplicative Logit Adjustment Approximates Neural-Collapse-Aware Decision Boundary Adjustment (ICLR 2025 · similarity 0.87)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.87)
- Learning Label Shift Correction for Test-Agnostic Long-Tailed Recognition (ICML 2024 · similarity 0.86)
- Harnessing Hierarchical Label Distribution Variations in Test Agnostic Long-tail Recognition (ICML 2024 · similarity 0.86)
- Rethinking the Bias of Foundation Model under Long-tailed Distribution (ICML 2025 · 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.