Research map / Reinforcement learning / MDPs and planning
Mitigating Bias in Face Recognition Using Skewness-Aware Reinforcement Learning
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use reinforcement-learning-selected adaptive margins instead of a uniform Softmax margin | Softmax-based face recognition · RFW (BUPT-Globalface) | average accuracy: 92.10 -> 94.45 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of a uniform Softmax margin | Softmax-based face recognition · RFW (BUPT-Globalface) | STD: 2.48 -> 0.44 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of Cosface | Cosface-based face recognition · RFW (BUPT-Globalface) | average accuracy: 94.25 -> 95.01 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of Arcface | Arcface-based face recognition · RFW (BUPT-Globalface) | average accuracy: 95.37 -> 95.79 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of a uniform Softmax margin | Softmax-based face recognition · RFW (BUPT-Balancedface) | STD: 1.38 -> 0.20 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of Cosface | Cosface-based face recognition · RFW (BUPT-Balancedface) | average accuracy: 93.74 -> 95.10 | Helped |
| Use reinforcement-learning-selected adaptive margins instead of Arcface | Arcface-based face recognition · RFW (BUPT-Balancedface) | average accuracy: 94.64 -> 95.19 | Helped |
| Replace reinforcement-learning-selected margins with fixed race-specific margins inversely proportional to sample counts (ablation) | Race-balance network with Softmax loss · RFW (BUPT-Globalface) | STD: 1.90 -> 0.44 | Helped |
| Use balanced rather than biased racial training data | Face recognition models · RFW | fairness | Helped |
| Train with a uniform margin across races, rather than adapting margins by race and training step | Large-margin face recognition · RFW | race balance | Hurt |
Closest papers by meaning
- Leveraging Diffusion Perturbations for Measuring Fairness in Computer Vision (AAAI 2024 · similarity 0.85)
- Classes Are Not Equal: An Empirical Study on Image Recognition Fairness (CVPR 2024 · similarity 0.84)
- Some Optimizers are More Equal: Understanding the Role of Optimizers in Group Fairness (NeurIPS 2025 · similarity 0.84)
- FAIR-FER: A Latent Alignment Approach for Mitigating Bias in Facial Expression Recognition (Student Abstract) (AAAI 2024 · similarity 0.84)
- Interpretable Debiasing of Vision-Language Models for Social Fairness (CVPR 2026 · similarity 0.83)
- On the Fairness ROAD: Robust Optimization for Adversarial Debiasing (ICLR 2024 · similarity 0.83)
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.