Research map / Learning theory and algorithms / Fairness
Metrics for Dataset Demographic Bias: A Case Study on Facial Expression Recognition
Iris Dominguez-Catena, Daniel Paternain, Mikel Galar
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 |
|---|---|---|---|
| remove all female subjects from training data (representational bias) (ablation) | ResNet50 on AffectNet · AffectNet gender Exp1 | angry recall, female group: 20% recall change between extremes | Hurt |
| remove one gender entirely from training data (ablation) | ResNet50 on AffectNet · AffectNet gender Exp1 | recall other classes: variations up to 10% | Hurt |
| remove female examples from happy class (ablation) | ResNet50 on AffectNet · AffectNet gender Exp2 | female happy recall: 0.85 -> 0.4 | Hurt |
| bias happy class while other classes stay balanced (ablation) | ResNet50 on AffectNet · AffectNet gender Exp2 | recall of unperturbed classes | No clear effect |
| bias happy and angry classes in opposite directions (ablation) | ResNet50 on AffectNet · AffectNet gender Exp3 | angry recall: 0.4 -> 0.1 | Hurt |
| give group all angry examples (opposite bias) (ablation) | ResNet50 on AffectNet · AffectNet gender Exp3 | angry recall: 0.4 -> 0.6 | Improved |
| bias two classes in opposite directions instead of one (ablation) | ResNet50 on AffectNet · AffectNet gender Exp3 | recall | No clear effect |
| use BP instead of IR^-1 for dominance | Representational bias metrics · FER datasets | Spearman agreement: 0.85 to 0.96 | Improved |
| use Z instead of NPMI for local bias | Local stereotypical bias metrics · FER+ age | interpretability: -1 and -0.22 vs -1 and -0.9 | Improved |
| use Cramer's V vs other chi2/NMI metrics | Stereotypical bias metrics · FER datasets | Spearman agreement: minimum 0.91 | No clear effect |
Closest papers by meaning
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.89)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (AAAI 2024 · similarity 0.89)
- Understanding and Mitigating Annotation Bias in Facial Expression Recognition (ICCV 2021 · similarity 0.88)
- Towards causal benchmarking of bias in face analysis algorithms (ECCV 2020 · similarity 0.87)
- AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (CVPR 2025 · similarity 0.86)
- Diverse Yet Biased: Towards Mitigating Biases in Generative AI (Student Abstract) (AAAI 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.
