Wasserstein Fair Classification
arXiv:1907.12059
Abstract
We propose an approach to fair classification that enforces independence between the classifier outputs and sensitive information by minimizing Wasserstein-1 distances. The approach has desirable theoretical properties and is robust to specific choices of the threshold used to obtain class predictions from model outputs. We introduce different methods that enable hiding sensitive information at test time or have a simple and fast implementation. We show empirical performance against different fairness baselines on several benchmark fairness datasets.
References in corpus (6)
- Scikit-learn: Machine Learning in Python
- Equality of Opportunity in Supervised Learning
- Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations
- Empirical Risk Minimization under Fairness Constraints
- Obtaining fairness using optimal transport theory
- Scalable Learning of Non-Decomposable Objectives
Cited by in corpus (17)
- Fairness in Machine Learning
- Training individually fair ML models with Sensitive Subspace Robustness
- Reducing Sentiment Bias in Language Models via Counterfactual Evaluation
- Review of Mathematical frameworks for Fairness in Machine Learning
- A case for new neural network smoothness constraints
- All of the Fairness for Edge Prediction with Optimal Transport
- Differentiable Ranks and Sorting using Optimal Transport
- Fairness guarantee in multi-class classification
- Modeling Techniques for Machine Learning Fairness: A Survey
- Understanding and Mitigating Accuracy Disparity in Regression
- Learning to Score Behaviors for Guided Policy Optimization
- A Consistent Extension of Discrete Optimal Transport Maps for Machine Learning Applications
- Fairness with Continuous Optimal Transport
- Towards Return Parity in Markov Decision Processes
- Unified Shapley Framework to Explain Prediction Drift
- Fairness in Deep Learning: A Computational Perspective
- Optimal Pre-Processing to Achieve Fairness and Its Relationship with Total Variation Barycenter