Research map / Label-efficient and robust learning / Recognition and long-tail
Learning from Failure: De-biasing Classifier from Biased Classifier
Junhyun Nam, Hyuntak Cha, Sungsoo Ahn, Jaeho Lee, Jinwoo Shin
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 |
|---|---|---|---|
| LfF reweighting by biased-model relative difficulty | Vanilla classifier · Colored MNIST 99.5% | unbiased accuracy: 35.34 -> 63.39 | Improved |
| LfF reweighting | Vanilla classifier · Corrupted CIFAR-10^1 99% | unbiased accuracy: 22.72 -> 41.37 | Improved |
| LfF (no bias labels) vs Group DRO (bias labels) | debiasing · Corrupted CIFAR-10^2 99.5% | unbiased accuracy: 34.25 -> 34.11 | No clear effect |
| LfF vs Group DRO | debiasing · CelebA | unbiased accuracy: HairColor 85.43 -> 84.24; HeavyMakeup 64.88 -> 66.20 | Mixed |
| LfF vs ReBias | debiasing · BAR Diving | accuracy: 51.57 -> 34.59 | Hurt |
| train biased model with GCE instead of CE (ablation) | LfF · Colored MNIST 99% | unbiased accuracy: 49.90 -> 74.01 | Improved |
| train biased model with GCE instead of CE (ablation) | LfF · Corrupted CIFAR-10^1 99% | unbiased accuracy: 25.15 -> 41.37 | Improved |
| swap target/bias so bias is harder than target | Vanilla classifier · Colored MNIST | unbiased accuracy: -0.03% | No clear effect |
Closest papers by meaning
- Learning Debiased Classifier with Biased Committee (NeurIPS 2022 · similarity 0.90)
- Self-supervised Debiasing Using Low Rank Regularization (CVPR 2024 · similarity 0.88)
- DeNetDM: Debiasing by Network Depth Modulation (NeurIPS 2024 · similarity 0.87)
- SEBRA : Debiasing through Self-Guided Bias Ranking (ICLR 2025 · similarity 0.87)
- Enhancing Intrinsic Features for Debiasing via Investigating Class-Discerning Common Attributes in Bias-Contrastive Pair (CVPR 2024 · similarity 0.86)
- Navigate Beyond Shortcuts: Debiased Learning Through the Lens of Neural Collapse (CVPR 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.
