Research map / Label-efficient and robust learning / Recognition and long-tail
Gaussian Affinity for Max-Margin Class Imbalanced Learning
Munawar Hayat, Salman Khan, Syed Waqas Zamir, Jianbing Shen, Ling Shao
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 cost-sensitive CNN with affinity loss | CoSen CNN · DIL skin lesion 5 classes | accuracy: 80.2 -> 91.1 | Improved |
| replace cost-sensitive CNN with affinity loss | CoSen CNN · DIL skin lesion 10 classes | accuracy: 72.6 -> 80.3 | Improved |
| replace CoSen CNN with affinity loss | CoSen CNN · Imbalanced MNIST (10% of even digits) | accuracy: 98.4 -> 99.3 | Improved |
| replace CoSen CNN with affinity loss | CoSen CNN · Imbalanced MNIST (25% of even digits) | accuracy: 98.5 -> 99.5 | Improved |
| replace CoSen CNN with affinity loss | CoSen CNN · Imbalanced MNIST (standard split) | accuracy: 99.3 -> 99.6 | Improved |
| replace soft-max loss with affinity loss under label noise | soft-max · MNIST with 10-50% flipped labels | error rate | Improved |
| replace soft-max with affinity loss across imbalance levels | soft-max · MNIST imbalance up to 1:0.025 | error rate | Improved |
| use at least 5 prototypes per class instead of fewer (ablation) | Affinity Loss · Imbalanced MNIST (10% of first five digits) | error rate: best for m >= 5, m <= 5 relatively poor | Improved |
| increase prototypes per class beyond 5 (ablation) | Affinity Loss · Imbalanced MNIST (10% of first five digits) | error rate | No clear effect |
| set sigma between 5 and 20 instead of very high values (ablation) | Affinity Loss · Imbalanced MNIST (10% of first five digits) | error rate: optimal sigma between 5 and 20 | Improved |
| use inverse distance similarity instead of Gaussian with squared L2 (ablation) | Affinity Loss · Imbalanced MNIST (10% of first five digits) | accuracy: 99.3 -> 99.1 | Hurt |
| use L1 distance with Gaussian similarity (ablation) | Affinity Loss · Imbalanced MNIST (10% of first five digits) | accuracy: fails to converge | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CosFace — CosFace: Large Margin Cosine Loss for Deep Face Recognition (CVPR 2018) · better on 2 of 2 shared results
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
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.90)
- Deep Imbalanced Learning for Face Recognition and Attribute Prediction (TPAMI 2019 · similarity 0.88)
- Influence-Balanced Loss for Imbalanced Visual Classification (ICCV 2021 · similarity 0.87)
- Learning Deep Representation for Imbalanced Classification (CVPR 2016 · similarity 0.87)
- Difficulty-aware Balancing Margin Loss for Long-tailed Recognition (AAAI 2025 · similarity 0.87)
- Robust Visual Recognition with Class-Imbalanced Open-World Noisy Data (AAAI 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.
