Research map / Probabilistic and scientific modeling / Variational and generative sampling
Group Fairness by Probabilistic Modeling with Latent Fair Decisions
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 |
|---|---|---|---|
| Model the observed label as a biased proxy of a latent fair decision, rather than enforcing parity directly on the observed label | Probabilistic fair classification · Synthetic data | accuracy on fair labels: around 0.17 higher than that of FairLR | Helped |
| Use probabilistic circuits instead of naive Bayes structure | Fair probabilistic modeling · Synthetic data | test log-likelihood: -7.361 (LatNB) vs -7.022 (FairPC-prior), 10 non-sensitive variables | Helped |
| Use a latent decision variable instead of enforcing fairness directly on the observed label | Probabilistic fair models · Real-world datasets | log-likelihood: COMPAS: -3.922 -> -3.919; Adult: -5.980 -> -5.962; German: -11.454 -> -11.422 | Helped |
| Add prior initialization encoding D=D_f instead of random parameter initialization (ablation) | FairPC · Synthetic data | test log-likelihood: -7.024 -> -7.022 | No clear effect |
| Add prior initialization encoding D=D_f instead of random parameter initialization (ablation) | FairPC · Synthetic data | convergence to the bias mechanism | Helped |
| Use a latent fair decision | Fair probabilistic models · Synthetic data | discrimination | Helped |
| Use FairPC rather than fair logistic regression methods | Fair classification · German | accuracy on observed labels | Mixed |
Closest papers by meaning
- Can I Trust My Fairness Metric? Assessing Fairness with Unlabeled Data and Bayesian Inference (NeurIPS 2020 · similarity 0.90)
- Certifying Fairness of Probabilistic Circuits (AAAI 2023 · similarity 0.89)
- Fairness without Harm: An Influence-Guided Active Sampling Approach (NeurIPS 2024 · similarity 0.89)
- Fairness without Demographics through Shared Latent Space-Based Debiasing (AAAI 2024 · similarity 0.89)
- Recycling Privileged Learning and Distribution Matching for Fairness (NeurIPS 2017 · similarity 0.89)
- On Group Sufficiency Under Label Bias (NeurIPS 2025 · similarity 0.89)
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.