[论文解读] Classification with Fairness Constraints: A Meta-Algorithm with Provable Guarantees

L. Elisa Celis, Lingxiao Huang+2 more|arXiv (Cornell University)|Jun 15, 2018

Ethics and Social Impacts of AI参考文献 42被引用 29

一句话总结

本文提出了一种新颖的元算法用于公平分类，可将广泛的公平性约束——包括预测均等性、统计均等性以及等化奇偶性——转化为凸优化问题，从而实现可证明的公平性保证。该方法将复杂的公平性约束简化为广义群体公平性框架，在多个数据集和公平性度量下实现了近乎完美的公平性，且准确率损失最小。

ABSTRACT

Developing classification algorithms that are fair with respect to sensitive attributes of the data has become an important problem due to the growing deployment of classification algorithms in various social contexts. Several recent works have focused on fairness with respect to a specific metric, modeled the corresponding fair classification problem as a constrained optimization problem, and developed tailored algorithms to solve them. Despite this, there still remain important metrics for which we do not have fair classifiers and many of the aforementioned algorithms do not come with theoretical guarantees; perhaps because the resulting optimization problem is non-convex. The main contribution of this paper is a new meta-algorithm for classification that takes as input a large class of fairness constraints, with respect to multiple non-disjoint sensitive attributes, and which comes with provable guarantees. This is achieved by first developing a meta-algorithm for a large family of classification problems with convex constraints, and then showing that classification problems with general types of fairness constraints can be reduced to those in this family. We present empirical results that show that our algorithm can achieve near-perfect fairness with respect to various fairness metrics, and that the loss in accuracy due to the imposed fairness constraints is often small. Overall, this work unifies several prior works on fair classification, presents a practical algorithm with theoretical guarantees, and can handle fairness metrics that were previously not possible.

研究动机与目标

• 解决现有公平分类算法在非凸公平性约束下缺乏可证明公平性保证的问题。
• 在单一算法框架下统一多种公平性定义，如预测均等性、统计均等性和等化奇偶性。
• 使原本因非凸优化而难以处理的公平性约束（尤其是假发现率和假排除率等度量）得以应用。
• 提供关于公平性与准确率权衡的理论保证，确保分类器满足用户定义的公平性阈值。
• 通过处理多个非互斥的敏感属性以及超越二元群体公平性的通用公平性度量，扩展先前工作。

提出的方法

• 将公平分类问题建模为带有公平性约束的约束优化任务，其中公平性约束通过敏感群体上的性能函数进行建模。
• 通过一种新颖的公平性度量转换方法，将一般公平性约束（如预测均等性）约化为凸的群体公平性约束族。
• 开发一种基于凸松弛和对偶性的核心算法，用于群体公平分类，实现高效优化并具备理论收敛保证。
• 应用一种元算法框架，将任意公平性度量（如假发现率、假排除率）通过参数化性能函数映射到凸群体公平框架。
• 使用线性分式性能函数来建模如预测均等性等公平性度量，从而可应用凸优化技术。
• 通过广义公平性函数和迭代优化，将框架扩展至同时处理多个敏感属性和多个公平性约束。

实验结果

研究问题

• RQ1 能否设计一种单一元算法，以可证明的理论保证处理广泛类别的公平性约束？
• RQ2 如何将非凸公平性约束（如预测均等性）约化为凸优化问题？
• RQ3 所提出的算法在多种公平性度量下，能在多大程度上实现高公平性且最小化准确率下降？
• RQ4 该算法在真实世界数据集（如COMPAS和Heart）上的表现如何，特别是在假发现率和假排除率等度量下？
• RQ5 该框架能否扩展以同时处理多个重叠的敏感属性和多个公平性约束？

主要发现

• 在FICO COMPAS数据集上，该算法实现了近乎完美的公平性（例如，γ_FDR ≈ 0.99），在公平性度量上优于SHIFT和COV等基线方法。
• 在假发现率（FDR）公平性方面，Algo 1 -FDR在COMPAS数据集上实现了最大公平性得分（γ_FDR）0.80，而SHIFT达到0.98，表明其对数据分布假设较为敏感。
• Algo 1 -SR在准确率仅下降6%（从0.83降至0.77）的情况下，实现了0.89的公平性得分（γ_SR），展现出优异的公平性-准确率权衡。
• 无约束最优分类器的准确率为83%，而Algo 1 -SR将其降至77%，但显著提升了多个度量的公平性，包括正预测率（γ_PPR）达到90%。
• Algo 1 -FDR在保持高准确率（0.83）的同时，实现了85%的假发现率公平性（γ_FDR），在公平性方面优于COV和FPR-COV基线，且准确率相当。
• 该框架成功解决了公平机器学习中的一个开放问题，实现了对预测均等性（包括假发现率和假排除率均等性）的可证明公平性保证，而此前的算法无法处理此类情况。