Achieving Fairness Without Harm via Selective Demographic Experts

2025-11-09

AI Summary

In high-stakes domains such as healthcare, achieving both fairness and predictive accuracy remains challenging. This paper proposes a “Selective Population Expert Mechanism” that learns group-specific representations and trains independent, personalized classifiers for each subgroup. Crucially, it dynamically activates the optimal expert per input under a harmlessness constraint—ensuring no subgroup’s performance degrades—thereby avoiding the accuracy trade-offs inherent in conventional debiasing methods. The framework is trained end-to-end via multi-task optimization, preserving original subgroup accuracy without compromise. Experiments on three real-world medical datasets (ocular disease, skin cancer, and X-ray diagnosis) and two facial attribute datasets demonstrate substantial fairness improvements—e.g., reductions of 37–62% in equal opportunity difference—while strictly maintaining baseline accuracy across all subgroups. To our knowledge, this is the first method to achieve simultaneous fairness enhancement and zero-accuracy-loss—a principled breakthrough in fair machine learning.

Technology Category

• Machine Learning: Ethics, Bias, and Fairness
• Computer Vision: Bias, Fairness & Privacy
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments

Abstract

As machine learning systems become increasingly integrated into human-centered domains such as healthcare, ensuring fairness while maintaining high predictive performance is critical. Existing bias mitigation techniques often impose a trade-off between fairness and accuracy, inadvertently degrading performance for certain demographic groups. In high-stakes domains like clinical diagnosis, such trade-offs are ethically and practically unacceptable. In this study, we propose a fairness-without-harm approach by learning distinct representations for different demographic groups and selectively applying demographic experts consisting of group-specific representations and personalized classifiers through a no-harm constrained selection. We evaluate our approach on three real-world medical datasets -- covering eye disease, skin cancer, and X-ray diagnosis -- as well as two face datasets. Extensive empirical results demonstrate the effectiveness of our approach in achieving fairness without harm.

Problem

• Ensuring fairness without compromising predictive performance in machine learning
• Addressing trade-offs between fairness and accuracy in bias mitigation
• Applying demographic-specific experts to maintain performance across groups

Innovation

• Learning distinct representations for demographic groups
• Selectively applying demographic experts with classifiers
• Using no-harm constrained selection for fairness