Does enforcing fairness mitigate biases caused by subpopulation shift?

Abstract

Many instances of algorithmic bias are caused by subpopulation shifts. For example, ML models often perform worse on demographic groups that are underrepresented in the training data. In this paper, we study whether enforcing algorithmic fairness during training improves the performance of the trained model in the \emph{target domain}. On one hand, we conceive scenarios in which enforcing fairness does not improve performance in the target domain. In fact, it may even harm performance. On the other hand, we derive necessary and sufficient conditions under which enforcing algorithmic fairness leads to the Bayes model in the target domain. We also illustrate the practical implications of our theoretical results in simulations and on real data.

References in corpus (11)

• Equality of Opportunity in Supervised Learning
• The Measure and Mismeasure of Fairness
• WILDS: A Benchmark of in-the-Wild Distribution Shifts
• No Classification without Representation: Assessing Geodiversity Issues in Open Data Sets for the Developing World
• Predictive Inequity in Object Detection
• A Reductions Approach to Fair Classification
• Training individually fair ML models with Sensitive Subspace Robustness
• Fairness Through Awareness
• Avoiding Disparity Amplification under Different Worldviews
• Certifying and removing disparate impact
• Training Well-Generalizing Classifiers for Fairness Metrics and Other Data-Dependent Constraints

Cited by in corpus (3)

• Toward Operationalizing Pipeline-aware ML Fairness: A Research Agenda for Developing Practical Guidelines and Tools
• Learning to Estimate Without Bias
• Fairness in Forecasting of Observations of Linear Dynamical Systems