The cost of fairness in binary classification

By Aditya Krishna Menon, Robert C. Williamson. 2018

491 citations recorded by the source corpus.

Subfields: Optimization & Training, 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 Optimization & Training

In AI Ethics, Fairness & Society

Citations in the corpus

Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “[108] note that adding Y (c− 1) and negating gives the cost-sensitive risk that Menon and Williamson aim to minimize [117]:”
Stable and Fair Classification (2019, methodology): “Given a classiﬁer f , we consider a fairness metric for statistical rate, which has been applied in [55, 4].”
Taking Advantage of Multitask Learning for Fair Classification (2018, background): “In recent years there has been a lot of interest in the problem of enhancing learning methods with “fairness” requirements, see [1– 4 8–10 12 17 19 20 22–26 28 30 31 35–39] and references therein.”
Characterizing Fairness Over the Set of Good Models Under Selective Labels (2021, background): “An influential literature on fair classification and fair regression constructs prediction functions that minimize loss subject to a predictive fairness constraint chosen by the decision maker [25, 9, 26, 27, 28, 10, 11, 29].”
Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classification (2019, background): “s of fairness and unintended bias apply to machine learning models. Much of this work provides metrics to quantify the presence of unintended bias according to specific definitions. In [8], [12], and [15], one can read overviews of several of these metrics and the relationship and trade-offs between”
Empirical Risk Minimization under Fairness Constraints (2018, background): “Methods in the second family enforce fairness directly during the training step: [18, 19, 20, 3, 21, 22, 23, 24, 25, 26] are examples of this method which provide non-convex approaches to the solution of the problem or they derive consistency results just for the non-convex formulation resorting…”
Algorithmic Fairness (2020, background): “[39] and [101] similarly suggest selecting separate thresholds for each group separately, in a manner that maximizes accuracy and minimizes demographic parity.”
Controllable Guarantees for Fair Outcomes via Contrastive Information Estimation (2021, background): “Therefore, it is desirable to produce representations that can trade-off some measure of fairness (e.g., statistical parity in this work) with utility (Menon and Williamson 2018; Dutta et al. 2019).”
Does mitigating ML's impact disparity require treatment disparity? (2017, background): “One can show that u(d, c) = −CS(d; c)+π(1− c), and hence the problem of maximizing immediate utility considered here is equivalent to minimizing cost-sensitive risk as in [21].”
Are My Deep Learning Systems Fair? An Empirical Study of Fixed-Seed Training (2021, background): “In addition, unlike accuracy, which is a widely-accepted metric, dozens of fairness metrics exist [29, 35, 46, 60, 64, 99, 108].”
Fairness Constraints: A Flexible Approach for Fair Classification (2019, background): “The third and final strategy consists of post-processing the classifier scores such that the new outcomes contain no disparate impact or disparate mistreatment (Hardt et al., 2016; Corbett-Davies et al., 2017; Dwork et al., 2018; Menon and Williamson, 2018).”
Perturbation Augmentation for Fairer NLP (2022, background): “…that the agreement (or lack thereof) of different NLP bias measurements warrants further exploration, and closer examinations of fairness evaluation datasets. performance (Zliobaite, 2015; Menon and Williamson, 2018) or seeking a Pareto optimal solution (Berk et al., 2017; Zhao and Gordon, 2019).”
Classification with Fairness Constraints: A Meta-Algorithm with Provable Guarantees (2018, methodology): “2 (Lemma 9 of [48]).”
Awareness in practice: tensions in access to sensitive attribute data for antidiscrimination (2019, background): “nting fairnessenhancing interventions, fair ML research and toolkits [9, 38, 67] often take for granted when framing problems and their solutions that sensitive attribute data are available as inputs [17, 32, 39, 40, 48, 50, 65]. When labeled data are not available, researchers have made powerful di”