TowerDebias: A Novel Unfairness Removal Method Based on the Tower Property

2024-11-13

AI Summary

To address prediction unfairness stemming from sensitive attributes (e.g., race, gender) in black-box models, this paper proposes a retraining-free post-processing debiasing method. The core method leverages the tower property of conditional expectation—introduced here for the first time in fairness correction—to construct a principled probabilistic framework grounded in conditional expectations; it formally establishes a fairness improvement theorem, ensuring cross-task generalizability without access to or assumptions about model internals. The approach unifies treatment of both classification and regression tasks. Extensive experiments on multiple real-world datasets demonstrate that inter-group disparity metrics—including ΔDP and ΔEO—are reduced below 0.02, while accuracy degradation remains within 1.5%, substantially outperforming existing post-processing baselines.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Bias, Fairness & Privacy
• Reasoning under Uncertainty: Relational Probabilistic Models

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

Decision-making processes have increasingly come to rely on sophisticated machine learning tools, raising critical concerns about the fairness of their predictions with respect to sensitive groups. The widespread adoption of commercial"black-box"models necessitates careful consideration of their legal and ethical implications for consumers. When users interact with such black-box models, a key challenge arises: how can the influence of sensitive attributes, such as race or gender, be mitigated or removed from its predictions? We propose towerDebias (tDB), a novel post-processing method designed to reduce the influence of sensitive attributes in predictions made by black-box models. Our tDB approach leverages the Tower Property from probability theory to improve prediction fairness without requiring retraining of the original model. This method is highly versatile, as it requires no prior knowledge of the original algorithm's internal structure and is adaptable to a diverse range of applications. We present a formal fairness improvement theorem for tDB and showcase its effectiveness in both regression and classification tasks using multiple real-world datasets.

Problem

• Mitigating influence of sensitive attributes in black-box model predictions
• Improving prediction fairness without retraining original models
• Removing bias from machine learning decisions ethically and legally

Innovation

• Uses Tower Property for unfairness removal
• Post-processing without retraining black-box models
• Versatile for various applications and datasets