Research map / Label-efficient and robust learning / Recognition and long-tail
Local and Global Logit Adjustments for Long-Tailed Learning
Yingfan Tao, Jingna Sun, Hao Yang, Li Chen, Xu Wang, Wenming Yang, Daniel Du, Min Zheng
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 |
|---|---|---|---|
| replace BALMS logit adjustment with Class-aware Logit Adjustment (CLA) (ablation) | 3-expert model · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 55.24 -> 56.53 | Improved |
| add AAW loss to BALMS (ablation) | 3-expert BALMS · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 55.24 -> 56.51 | Improved |
| add AAW loss to CLA (ablation) | CLA 3-expert model · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 56.53 -> 57.15 | Improved |
| use RandAugment (ablation) | Softmax baseline · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 41.88 -> 47.97 | Improved |
| use RandAugment (ablation) | LGLA (GC) · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 49.58 -> 57.15 | Improved |
| increase number of experts beyond 3 (ablation) | LGLA · CIFAR-100-LT (IF=100) | top-1 accuracy (%) | No clear effect |
| increase number of experts from 2 to 3 (ablation) | LGLA · CIFAR-100-LT (IF=100) | top-1 accuracy (%) | Improved |
| use group convolution instead of independent convolution for experts (ablation) | LGLA · CIFAR-100-LT (IF=100) | top-1 accuracy (%): 56.5 -> 57.2 | Improved |
| raise fusion weight of an LLA expert (ablation) | 4-expert LGLA · CIFAR-100-LT (IF=100) | top-1 accuracy on matching split | Improved |
| LGLA vs ResLT/NCL on Many-shot classes | LGLA · iNaturalist 2018 (Many) | top-1 accuracy (%): 73.0 -> 69.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- ResLT — ResLT: Residual Learning for Long-tailed Recognition (TPAMI 2022) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Class and Attribute-Aware Logit Adjustment for Generalized Long-Tail Learning (AAAI 2025 · similarity 0.89)
- Robust Logit Adjustment for Learning with Long-Tailed Noisy Data (AAAI 2025 · similarity 0.86)
- Multiplicative Logit Adjustment Approximates Neural-Collapse-Aware Decision Boundary Adjustment (ICLR 2025 · similarity 0.84)
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.84)
- Breaking Long-Tailed Learning Bottlenecks: A Controllable Paradigm with Hypernetwork-Generated Diverse Experts (NeurIPS 2024 · similarity 0.84)
- Learning Only When It Matters: Cost-Aware Long-Tailed Classification (AAAI 2024 · similarity 0.83)
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.
