A Confidence-Based Approach for Balancing Fairness and Accuracy

By Benjamin Fish, Jeremy Kun, Á. Lelkes. 2016

265 citations recorded by the source corpus.

We study three classical machine learning algorithms in the context of algorithmic fairness: adaptive boosting, support vector machines, and logistic regression. Our goal is to maintain the high accuracy of these learning algorithms while reducing the degree to which they discriminate against individuals because of their membership in a protected group. Our first contribution is a method for achieving fairness by shifting the decision boundary for the protected group. The method is based on the theory of margins for boosting. Our method performs comparably to or outperforms previous algorithms in the fairness literature in terms of accuracy and low discrimination, while simultaneously allowing for a fast and transparent quantification of the trade-off between bias and error. Our second contribution addresses the shortcomings of the bias-error trade-off studied in most of the algorithmic fairness literature. We demonstrate that even hopelessly naive modifications of a biased algorithm, which cannot be reasonably said to be fair, can still achieve low bias and high accuracy. To help to distinguish between these naive algorithms and more sensible algorithms we propose a new measure of fairness, called resilience to random bias (RRB). We demonstrate that RRB distinguishes well between our naive and sensible fairness algorithms. RRB together with bias and accuracy provides a more complete picture of the fairness of an algorithm.

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

Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments (2016, background): “While our discussion focusses primarily on first three fairness criteria, statistical parity is widely used within the machine learning community and may be the criterion with which many readers are most familiar[16, 17].”
Learning Optimal and Fair Decision Trees for Non-Discriminative Decision-Making (2019, methodology): “Second, post-processing approaches, which a-posteriori adjust the predictors learned using standard ML techniques to improve their fairness properties (Hardt, Price, and Srebro 2016; Fish, Kun, and Lelkes 2016).”
Stable and Fair Classification (2019, background): “Another approach for fair classiﬁcation is to shift the decision boundary of a baseline classiﬁer, e.g., [28, 37, 34, 61, 73, 22].”
The Price of Fair PCA: One Extra Dimension (2018, background): “…attribute directly [Pedreshi et al., 2008; Kamiran et al., 2010; Calders and Verwer, 2010; Kamiran and Calders, 2011; Luong et al., 2011; Kamiran et al., 2012; Kamishima et al., 2012; Hajian and Domingo-Ferrer, 2013; Feldman et al., 2015; Zafar et al., 2015; Fish et al., 2016; Adler et al., 2016].”
Adaptive Sensitive Reweighting to Mitigate Bias in Fairness-aware Classification (2018, methodology): “baselines employed by [17, 44]) propose that weights in Eq.”
A Convex Framework for Fair Regression (2017, background): “…of data (e.g. [18]); inprocessing, during the training of a model as either a constraint or incorporated into the objective function (e.g. [11, 21, 33]); or postprocessing, where data is labeled by some black-box model and then relabeled as a function only of the original labels (e.g.…”
Algorithmic Decision Making and the Cost of Fairness (2017, background): “(1) Statistical parity means that an equal proportion of defen-dants are detained in each race group [16, 17, 27, 38].”
Fairness in Learning: Classic and Contextual Bandits (2016, background): “…of outcomes , statistical parity – or algorithms that avoid disparate impact (see e.g. Calders and Verwer [2010], Luong et al. [2011], Kamishima et al. [2011], Feldman et al. [2015], Fish et al. [2016] and Adler et al. [2016] for a study of auditing existing algorithms for disparate impact).”
On the (im)possibility of fairness (2016, background): “[11] consider a machine learning algorithm to be fair if it can reconstruct the original labels of training data when noise has been added to the labels for anyone from a given protected class.”
Calibration for the (Computationally-Identifiable) Masses (2017, positive): “This result adds to the growing literature on fairness-accuracy trade-oﬀs [FKL16,BHJ + 17,CG17].”
Optimization with Non-Differentiable Constraints with Applications to Fairness, Recall, Churn, and Other Goals (2018, background): “…required to give loans at equal rates to different groups to alleviate disparity (Bocian et al., 2008), that is, the classifier is required to provide equal positive rates of classification across groups (see e.g. Zafar et al. (2015); Fish et al. (2016); Hardt et al. (2016); Goh et al. (2016)).”
Fair Learning in Markovian Environments (2016, background): “The second relaxation of fairness, approximate-action fairness, requires that an algorithm never favor an action of substantially lower quality than that of a better action.”
Proxy Fairness (2018, background): “Statistical Parity [20, 6, 11, 7] Pr { ˆ Y = 1 | G k = 1 } = Pr { ˆ Y = 1 } ∀ k Equal Opportunity [11] Pr { ˆ Y = 1 | Y = 1 , G k = 1 } = Pr { ˆ Y = 1 | Y = 1 } ∀ k Equal Odds [11] Pr { a given group, and across multiple groups, as well as conﬂict with the usual goal of maximizing classiﬁer…”
Why Is My Classifier Discriminatory? (2018, background): “In particular, the choice of hypothesis class H and learning objective has received a lot of attention (Calders & Verwer, 2010; Zemel et al., 2013; Fish et al., 2016).”
Classification with Fairness Constraints: A Meta-Algorithm with Provable Guarantees (2018, background): “There are increasingly many works on fairness in machine learning with provable guarantees, including [25, 34, 54, 57], that provide different classification algorithms with constraints on statistical parity or equalized odds, and [36, 40] for fairness in multi-armed bandit settings or ranking probl”
On Ensuring that Intelligent Machines Are Well-Behaved (2017, background): “…and adaptations of this work [Calders and Verwer, 2010; Luong et al., 2011; Kamishima et al., 2011; Dwork et al., 2012; Feldman et al., 2015; Fish et al., 2016; Joseph et al., 2016] and a long history of related game-theoretic efforts [Rabin, 1993; Fehr and Schmidt, 1999; Falk and…”
Your fairness may vary: Pretrained language model fairness in toxic text classification (2021, background): “…a relaxed nearly-optimal procedure for optimizing equalized odds (Woodworth et al., 2017), shifting the decision boundary for the protected group (Fish et al., 2016), iterative post-processing to achieve unbiased predictions on every identiﬁable subpopulation (Kim et al., 2019), recalibrating…”
Multicalibration: Calibration for the (Computationally-Identifiable) Masses (2018, result): “This result adds to the growing literature on fairness-accuracy trade-offs (Fish et al., 2016; Berk et al., 2017; Chouldechova & G’Sell, 2017).”
Fairness in Machine Learning (2020, methodology): “Over the last few years, researchers have introduced a rich set of definitions formalizing different fairness desiderata that can be used for evaluating and designing ML systems [2, 7, 14, 15, 21, 22, 23, 25, 28, 29, 31, 34, 44, 45, 46, 52, 53, 54, 55, 56, 59, 66, 69, 71, 72, 73, 74, 75, 82, 84, 87,”