Research map / Label-efficient and robust learning / Recognition and long-tail
Controllable Feature Whitening for Hyperparameter-Free Bias Mitigation
Yooshin Cho, Hanbyel Cho, Janghyeon Lee, HyeongGwon Hong, Jaesung Ahn, Junmo Kim
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 |
|---|---|---|---|
| add CFW (lambda=0.25) (ablation) | Vanilla · bFFHQ | unbiased accuracy: 77.96 -> 83.9 | Improved |
| add CFW (lambda=0.25) (ablation) | Vanilla · bFFHQ | bias-conflicting accuracy: 56.87 -> 78.1 | Improved |
| add loss weighting to Vanilla network (ablation) | Vanilla · bFFHQ | unbiased accuracy: 77.96 -> 78.5 | No clear effect |
| add loss weighting on top of CFW lambda=0.25 (ablation) | CFW · bFFHQ | unbiased accuracy: 83.9 -> 84.6 | Improved |
| use unbiased covariance weighting lambda=0.25 instead of 0.0 (ablation) | CFW · bFFHQ | unbiased accuracy: 80.0 -> 83.9 | Improved |
| use lambda=0.25 instead of 0.0 (ablation) | CFW · bFFHQ | bias-conflicting accuracy: 80.6 -> 78.1 | Hurt |
| use coupled Newton-Schulz T=5 instead of ZCA (ablation) | CFW · bFFHQ | bias-conflicting accuracy: 72.66 -> 79.8 | Improved |
| use cNS T=5 instead of Cholesky decomposition (ablation) | CFW · bFFHQ | unbiased accuracy: 74.06 -> 84.6 | Improved |
| use cNS T=5 instead of T=3 (ablation) | CFW · bFFHQ | bias-conflicting accuracy: 76.80 -> 79.8 | Improved |
| use SelecMix pretrained target encoder instead of Vanilla (ablation) | Ours · bFFHQ | bias-conflicting accuracy: 79.8 -> 82.77 | Improved |
| use SelecMix pretrained target encoder instead of Vanilla (ablation) | Ours · Corrupted CIFAR-10 0.5% | unbiased accuracy: 32.08 -> 42.51 | Improved |
| use CFW instead of adversarial GRL | Ours · Celeb-A T=attractive/S=male | equalized odds violation: 24.9 -> 4.2 | Improved |
Closest papers by meaning
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.85)
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.83)
- From Hope to Safety: Unlearning Biases of Deep Models via Gradient Penalization in Latent Space (AAAI 2024 · similarity 0.83)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.83)
- MABR: Multilayer Adversarial Bias Removal Without Prior Bias Knowledge (AAAI 2025 · similarity 0.83)
- Controlling for Omitted Variable Bias in Deep Neural Networks (NeurIPS 2026 · similarity 0.82)
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.
