Research map / Federated learning and privacy / Differential privacy
Differentially Private and Fair Deep Learning: A Lagrangian Dual Approach
Read the paperCheck 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 |
|---|---|---|---|
| Train PF-LD with demographic-parity constraints and sensitive-attribute privacy | PF-LD · Income | fairness violation: 0.181 -> 0.019 | Helped |
| Train PF-LD with accuracy-parity constraints and sensitive-attribute privacy | PF-LD · Bank | accuracy: 0.8 -> 0.812 | Helped |
| Use PF-LD rather than the private randomized-response method M | PF-LD · all evaluated datasets and fairness definitions | accuracy or fairness violation: 90%(27/30) | Mixed |
| Increase privacy budget epsilon | PF-LD · privacy-fairness-accuracy tradeoff experiments | fairness violation | Helped |
| Increase privacy budget epsilon | PF-LD · privacy-fairness-accuracy tradeoff experiments | accuracy | Hurt |
| Use intermediate primal clipping values C_p in [10, 20] rather than very small or large values (ablation) | PF-LD · Bank and Income | accuracy and fairness tradeoff | Helped |
| Use very small primal clipping values C_p (ablation) | PF-LD · primal clipping experiments | fairness violation | Hurt |
| Use very large primal clipping values C_p (ablation) | PF-LD · primal clipping experiments | accuracy | Hurt |
| Use large dual clipping values C_d (ablation) | PF-LD · Bank | accuracy | Hurt |
| Use small dual clipping values C_d (ablation) | PF-LD · Bank | fairness violation | Hurt |
| Reduce the fraction of training samples with reported sensitive attributes (ablation) | PF-LD · Bank, demographic parity | fairness violation | Hurt |
Closest papers by meaning
- Differentially Private Fair Learning (ICML 2019 · similarity 0.89)
- Training Private Models That Know What They Don’t Know (NeurIPS 2023 · similarity 0.88)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.88)
- Differential Privacy Has Disparate Impact on Model Accuracy (NeurIPS 2019 · similarity 0.88)
- Can Private Machine Learning Be Fair? (AAAI 2025 · similarity 0.86)
- Adversarial Fairness Network (AAAI 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.