Multi-Output Distributional Fairness via Post-Processing

2024-08-31

arXiv.org

AI Summary

Existing post-processing fairness methods lack generality for multi-output models (e.g., multi-task classification, representation learning), particularly for ensuring distributional parity across demographic groups. Method: This paper proposes the first generic post-processing framework targeting distributional parity. Its core innovation lies in integrating optimal transport mappings with Wasserstein barycenter theory for multi-output fairness calibration—introducing scalable barycenter approximation and kernel regression-based extrapolation to overcome the restrictive single-output assumption. The method operates via empirical distribution calibration and accommodates arbitrary multi-dimensional output structures. Results: Evaluated on multi-task classification and representation learning tasks, the framework significantly improves cross-group output distribution alignment. It achieves superior fairness across multiple metrics compared to state-of-the-art post-processing baselines while preserving original model accuracy without degradation.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Bias, Fairness & Privacy
• Game Theory and Economic Paradigms: Fair Division

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

The post-processing approaches are becoming prominent techniques to enhance machine learning models' fairness because of their intuitiveness, low computational cost, and excellent scalability. However, most existing post-processing methods are designed for task-specific fairness measures and are limited to single-output models. In this paper, we introduce a post-processing method for multi-output models, such as the ones used for multi-task/multi-class classification and representation learning, to enhance a model's distributional parity, a task-agnostic fairness measure. Existing methods for achieving distributional parity rely on the (inverse) cumulative density function of a model's output, restricting their applicability to single-output models. Extending previous works, we propose to employ optimal transport mappings to move a model's outputs across different groups towards their empirical Wasserstein barycenter. An approximation technique is applied to reduce the complexity of computing the exact barycenter and a kernel regression method is proposed to extend this process to out-of-sample data. Our empirical studies evaluate the proposed approach against various baselines on multi-task/multi-class classification and representation learning tasks, demonstrating the effectiveness of the proposed approach.

Problem

• Enhance fairness in multi-output machine learning models.
• Address limitations of single-output fairness post-processing methods.
• Propose optimal transport for distributional parity in multi-task models.

Innovation

• Post-processing for multi-output models
• Optimal transport mappings for fairness
• Kernel regression for out-of-sample data