Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Towards The Largest Margins
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 sample margin regularization to CE (ablation) | CE · CIFAR-10 | accuracy: 94.12 -> 94.45 | Helped |
| Add sample margin regularization to CE (ablation) | CE · CIFAR-100 | accuracy: 74.56 -> 74.96 | Helped |
| Add sample margin regularization to CE (ablation) | CE · MNIST | class margin: 87.39° -> 95.41° | Helped |
| Add zero-centroid regularization (ablation) | LM-Softmax · Imbalanced CIFAR-10, long-tailed, ratio 100 | accuracy: 65.32 -> 73.21 | Helped |
| Add zero-centroid regularization (ablation) | LM-Softmax · Imbalanced CIFAR-100, long-tailed, ratio 100 | accuracy: 41.52 -> 45.28 | Helped |
| Add zero-centroid regularization (ablation) | LM-Softmax · Imbalanced CIFAR-10, long-tailed, ratio 100 | class margin: 4.420° -> 92.57° | Helped |
| Add zero-centroid regularization (ablation) | NormFace · Imbalanced CIFAR-10, long-tailed, ratio 100 | accuracy: 72.37 -> 72.07 | Hurt |
| Use LM-Softmax instead of CE | person re-identification model · Market-1501 | mAP: 82.8 -> 84.7 | Helped |
| Use LM-Softmax instead of CE | person re-identification model · DukeMTMC | Rank@1: 83.5 -> 86.4 | Helped |
| Use LM-Softmax instead of ArcFace | face verification model · IJB-C | verification score: 99.4919 -> 99.4721 | Hurt |
| Add zero-centroid regularization (ablation) | LM-Softmax · IJB-C | verification score: 99.4721 -> 99.5086 | Helped |
| Use sample margin regularization as the sole loss rather than as a regularizer (ablation) | sample margin objective · CIFAR-100 | accuracy | Hurt |
Closest papers by meaning
- Large-Margin Softmax Loss for Convolutional Neural Networks (ICML 2016 · similarity 0.92)
- Mis-Classified Vector Guided Softmax Loss for Face Recognition (AAAI 2020 · similarity 0.92)
- Large Margin Deep Networks for Classification (NeurIPS 2018 · similarity 0.92)
- Reducing Class-Wise Performance Disparity via Margin Regularization (ICLR 2026 · similarity 0.90)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.87)
- Input Margins Can Predict Generalization Too (AAAI 2024 · 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.