Research map / Label-efficient and robust learning / Recognition and long-tail
Unbiased Classification through Bias-Contrastive and Bias-Balanced Learning
Youngkyu Hong, Eunho Yang
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 |
|---|---|---|---|
| BiasCon loss vs vanilla training | ResNet/CNN · BiasedMNIST rho=0.999 | unbiased accuracy: 11.8 -> 94.5 | Improved |
| add BiasBal loss to BiasCon (BC+BB) (ablation) | BiasCon · BiasedMNIST rho=0.995 | unbiased accuracy: 97.4 -> 97.7 | Improved |
| add BiasBal loss to BiasCon (BC+BB) (ablation) | BiasCon · BiasedMNIST rho=0.999 | unbiased accuracy: 94.5 -> 94.0 | Hurt |
| add BiasBal loss to BiasCon (BC+BB) (ablation) | BiasCon · CelebA Blonde | unbiased accuracy: 90.0 -> 91.4 | Improved |
| add BiasBal loss to BiasCon (BC+BB) (ablation) | BiasCon · CelebA Makeup | unbiased accuracy: 77.5 -> 78.6 | Improved |
| add BiasBal loss to BiasCon (BC+BB) (ablation) | BiasCon · UTKFace Age | unbiased accuracy: 75.7 -> 79.1 | Improved |
| BiasBal alone vs BiasCon alone (ablation) | BiasCon · CelebA Blonde | unbiased accuracy: 90.0 -> 91.1 | Mixed |
| BiasCon vs EnD (triplet-loss feature entangling) | EnD · BiasedMNIST rho=0.999 | unbiased accuracy: 59.5 -> 94.5 | Improved |
| SoftCon vs vanilla without bias labels | ResNet/CNN · BiasedMNIST rho=0.999 | unbiased accuracy: 11.8 -> 65.0 | Improved |
| SoftCon vs vanilla | ResNet18 · 9-Class ImageNet | unbiased accuracy: 92.7 -> 94.1 | Improved |
| SoftCon vs vanilla | ResNet18 · ImageNet-A | accuracy: 30.5 -> 34.1 | Improved |
| BiasBal alone vs vanilla (ablation) | CNN · BiasedMNIST rho=0.999 | unbiased accuracy: 11.8 -> 76.8 | Improved |
Closest papers by meaning
- Unbiased Supervised Contrastive Learning (ICLR 2023 · similarity 0.88)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.85)
- Enhancing Intrinsic Features for Debiasing via Investigating Class-Discerning Common Attributes in Bias-Contrastive Pair (CVPR 2024 · similarity 0.84)
- Classifier-to-Bias: Toward Unsupervised Automatic Bias Detection for Visual Classifiers (CVPR 2025 · similarity 0.84)
- CDMAD: Class-Distribution-Mismatch-Aware Debiasing for Class-Imbalanced Semi-Supervised Learning (CVPR 2024 · similarity 0.84)
- Self-supervised Debiasing Using Low Rank Regularization (CVPR 2024 · similarity 0.84)
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.
