To Split or Not to Split: The Impact of Disparate Treatment in Classification
arXiv:2002.04788
Abstract
Disparate treatment occurs when a machine learning model yields different decisions for individuals based on a sensitive attribute (e.g., age, sex). In domains where prediction accuracy is paramount, it could potentially be acceptable to fit a model which exhibits disparate treatment. To evaluate the effect of disparate treatment, we compare the performance of split classifiers (i.e., classifiers trained and deployed separately on each group) with group-blind classifiers (i.e., classifiers which do not use a sensitive attribute). We introduce the benefit-of-splitting for quantifying the performance improvement by splitting classifiers. Computing the benefit-of-splitting directly from its definition could be intractable since it involves solving optimization problems over an infinite-dimensional functional space. Under different performance measures, we (i) prove an equivalent expression for the benefit-of-splitting which can be efficiently computed by solving small-scale convex programs; (ii) provide sharp upper and lower bounds for the benefit-of-splitting which reveal precise conditions where a group-blind classifier will always suffer from a non-trivial performance gap from the split classifiers. In the finite sample regime, splitting is not necessarily beneficial and we provide data-dependent bounds to understand this effect. Finally, we validate our theoretical results through numerical experiments on both synthetic and real-world datasets.
References in corpus (28)
- Adam: A Method for Stochastic Optimization
- The information bottleneck method
- Equality of Opportunity in Supervised Learning
- OpenML: networked science in machine learning
- Counterfactual Fairness
- Estimating divergence functionals and the likelihood ratio by convex risk minimization
- Domain Adaptation: Learning Bounds and Algorithms
- The Measure and Mismeasure of Fairness
- MINE: Mutual Information Neural Estimation
- Fairness Testing: Testing Software for Discrimination
- Preventing Fairness Gerrymandering: Auditing and Learning for Subgroup Fairness
- -divergence Inequalities
- On Fairness and Calibration
- On the (im)possibility of fairness
- Fairness Without Demographics in Repeated Loss Minimization
- Learning Non-Discriminatory Predictors
- Local Privacy and Minimax Bounds: Sharp Rates for Probability Estimation
- A Framework for Understanding Sources of Harm throughout the Machine Learning Life Cycle
- Residual Unfairness in Fair Machine Learning from Prejudiced Data
- Does mitigating ML's impact disparity require treatment disparity?
- Identifying Significant Predictive Bias in Classifiers
- Estimating Skin Tone and Effects on Classification Performance in Dermatology Datasets
- Fairer and more accurate, but for whom?
- Inherent Tradeoffs in Learning Fair Representations
- Welfare and Distributional Impacts of Fair Classification
- Fairness With Minimal Harm: A Pareto-Optimal Approach For Healthcare
- Hypothesis Testing under Maximal Leakage Privacy Constraints
- Dualizing Le Cam's method for functional estimation, with applications to estimating the unseens