Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Disentangled Representation for Fair Facial Attribute Classification via Fairness-aware Information Alignment
Sungho Park, Sunhee Hwang, Dohyung Kim, Hyeran Byun
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 |
|---|---|---|---|
| disentangle representations into TAL, PAL, and MAL, align attribute information with decorrelation loss, and re-encode MAL for classification | FD-VAE · CelebA, attractiveness with male and young as protected attributes | equal opportunity: 1.3; 3.7 | Improved |
| disentangle representations into TAL, PAL, and MAL, align attribute information with decorrelation loss, and re-encode MAL for classification | FD-VAE · UTK Face, ethnicity and age with gender as protected attribute | equalized odds: 1.1; 2.9 | Improved |
| add decorrelation loss LDC to FFVAE (ablation) | FFVAE on CelebA · CelebA, attractiveness with male as protected attribute | equal opportunity: 15.0 -> 3.4 | Improved |
| add decorrelation loss LDC to FFVAE (ablation) | FFVAE on CelebA · CelebA, attractiveness with male as protected attribute | accuracy: 62.2 -> 64.1 | Improved |
| add MAL to the decorrelation-loss representation (ablation) | FD-VAE on CelebA · CelebA, attractiveness with male as protected attribute | equalized odds: 5.7 -> 2.1 | Improved |
| add MAL to the decorrelation-loss representation (ablation) | FD-VAE on CelebA · CelebA, attractiveness with male as protected attribute | accuracy: 64.1 -> 60.3 | Hurt |
| use MAL directly with TAL instead of re-encoding MAL (ablation) | FD-VAE downstream classifier · CelebA, attractiveness with male as protected attribute | accuracy: 60.3 -> 67.1 | Improved |
| use MAL directly with TAL instead of re-encoding MAL (ablation) | FD-VAE downstream classifier · CelebA, attractiveness with male as protected attribute | equal opportunity: 1.8 -> 19.8 | Hurt |
| re-encode MAL to retain target information while excluding protected information (ablation) | FD-VAE downstream classifier · CelebA, attractiveness with male as protected attribute | equal opportunity: 19.8 -> 1.3 | Improved |
| apply the downstream re-encoding framework to FFVAE (ablation) | FFVAE · CelebA, attractiveness with male as protected attribute | accuracy: 62.2 -> 63.5 | Improved |
| apply the downstream re-encoding framework to FFVAE (ablation) | FFVAE · CelebA, attractiveness with male as protected attribute | equal opportunity: 15.0 -> 15.3 | Hurt |
Closest papers by meaning
- Mitigating Demographic Bias in Facial Datasets with Style-Based Multi-attribute Transfer (IJCV 2021 · similarity 0.89)
- Fair Contrastive Learning for Facial Attribute Classification (CVPR 2022 · similarity 0.88)
- Distributionally Generative Augmentation for Fair Facial Attribute Classification (CVPR 2024 · similarity 0.86)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.86)
- FADES: Fair Disentanglement with Sensitive Relevance (CVPR 2024 · similarity 0.86)
- FairDD: Fair Dataset Distillation (NeurIPS 2025 · similarity 0.85)
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.
