Research map / Learning theory and algorithms / Fairness
Benchmarking Stochastic Approximation Algorithms for Fairness-Constrained Training of Deep Neural Networks
Andrii Kliachkin, Jana Lepšová, Gilles Bareilles, Jakub Marecek
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 |
|---|---|---|---|
| train with fairness constraints (SSL-ALM) instead of unconstrained SGD | MLP on ACSIncome · Oklahoma binary race, test | independence, separation, sufficiency: independence 0.09837 -> 0.08256; separation 0.15485 -> 0.10806; sufficiency 0.06096 -> 0.04598 | Improved |
| train with fairness constraints (SSL-ALM) instead of unconstrained SGD | MLP on ACSIncome · Oklahoma binary race, test | inaccuracy: 0.20878 -> 0.22321 | Hurt |
| use Stochastic Ghost constrained training | MLP on ACSIncome · Oklahoma binary race, test | inaccuracy: 0.20878 -> 0.23896 | Hurt |
| use Switching Subgradient constrained training | MLP on ACSIncome · Oklahoma binary race, test | fairness metrics: independence 0.09837 -> 0.10275 | No clear effect |
| add Fairret fairness regularizer to SGD | MLP on ACSIncome · Oklahoma binary race, test | independence, separation: independence 0.09837 -> 0.09064 | No clear effect |
| raise Fairret penalty multiplier from 0 to 3 (ablation) | SGD-Fairret · Oklahoma validation | loss vs constraint violation: loss 0.406 -> 0.539; violation 0.020 -> 0.004 | Mixed |
| use constrained algorithms instead of penalized SGD | MLP on ACSIncome · Virginia multi-valued attribute | loss and constraint satisfaction | No clear effect |
| use constrained algorithms instead of penalized SGD | MLP on ACSIncome · Virginia multi-valued attribute | hyperparameter sensitivity | Improved |
| remove smoothing term (ALM vs SSL-ALM) (ablation) | SSL-ALM · Oklahoma validation | loss: 0.4465 -> 0.4448 | No clear effect |
Closest papers by meaning
- Some Optimizers are More Equal: Understanding the Role of Optimizers in Group Fairness (NeurIPS 2025 · similarity 0.86)
- Enhancing Robust Fairness via Confusional Spectral Regularization (ICLR 2025 · similarity 0.86)
- Fair Bilevel Neural Network (FairBiNN): On Balancing fairness and accuracy via Stackelberg Equilibrium (NeurIPS 2024 · similarity 0.85)
- FFB: A Fair Fairness Benchmark for In-Processing Group Fairness Methods (ICLR 2024 · similarity 0.85)
- On the Fairness ROAD: Robust Optimization for Adversarial Debiasing (ICLR 2024 · similarity 0.85)
- ABCFair: an Adaptable Benchmark approach for Comparing Fairness Methods (NeurIPS 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.
