Fair Regression: Quantitative Definitions and Reduction-based Algorithms

By Alekh Agarwal, Miroslav Dud'ik, Zhiwei Steven Wu. 2019

308 citations recorded by the source corpus.

In this paper, we study the prediction of a real-valued target, such as a risk score or recidivism rate, while guaranteeing a quantitative notion of fairness with respect to a protected attribute such as gender or race. We call this class of problems \emph{fair regression}. We propose general schemes for fair regression under two notions of fairness: (1) statistical parity, which asks that the prediction be statistically independent of the protected attribute, and (2) bounded group loss, which asks that the prediction error restricted to any protected group remain below some pre-determined level. While we only study these two notions of fairness, our schemes are applicable to arbitrary Lipschitz-continuous losses, and so they encompass least-squares regression, logistic regression, quantile regression, and many other tasks. Our schemes only require access to standard risk minimization algorithms (such as standard classification or least-squares regression) while providing theoretical guarantees on the optimality and fairness of the obtained solutions. In addition to analyzing theoretical properties of our schemes, we empirically demonstrate their ability to uncover fairness--accuracy frontiers on several standard datasets.

Subfields: Safety Research, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Safety Research

In AI Ethics, Fairness & Society

Citations in the corpus

A Survey on Bias and Fairness in Machine Learning (2019, background): “In addition to the previous work, [1] considers the fair regression problem formulation with regards to two notions of fairness statistical (demographic) parity and bounded group loss.”
Characterizing Fairness Over the Set of Good Models Under Selective Labels (2021, methodology): “We construct discrete approximations to the loss function and disparity measure following the argument and notation in [23].”
Algorithmic Fairness (2020, methodology): “Note that other closely related problems are aimed at finding equilibrium (minimax) points in the field of game theory, and therefore, game-theoretic schemes may also be used for solutions [4, 5, 32, 55].”
Addressing bias in big data and AI for health care: A call for open science (2021, background): “To increase algorithmic fairness, protected attributes, such as gender or ethnicity, can be included during training in order to ensure that algorithmic predictions are statistically independent from these attributes.(34) Alternatively, loss functions can be defined per protected group, and may be f”
Debiasing Pre-Trained Language Models via Efficient Fine-Tuning (2022, methodology): “To mitigate biased behaviors in machine learning models, researchers have proposed meth-ods targeting different tasks and domains, such as classification (Menon and Williamson, 2018; Roh et al., 2021), regression (Agarwal et al., 2019; Berk et al., 2017), and adversarial learning (Xu et al., 2018).”
Fairness in Machine Learning: A Survey (2020, methodology): “Blinding (or partial blinding) has also been used as a fairness audit mechanism [2, 82, 143].”
A Review on Fairness in Machine Learning (2022, methodology): “One is an adjusted demographic parity measure, which requires the predictor to be independent of the sensitive attribute as measured by the cumulative distribution function (CDF) of the protected group compared to the CDF of the general population [5] using the Kolmogorov-Smirnov statistic”