Decoupled Classifiers for Group-Fair and Efficient Machine Learning

By Cynthia Dwork, Nicole Immorlica, Adam Tauman Kalai, Mark D. M. Leiserson. 2017

283 citations recorded by the source corpus.

Subfields: Efficiency & Compression, 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 Efficiency & Compression

In AI Ethics, Fairness & Society

Citations in the corpus

Stable and Fair Classification (2019, neutral): “, [30, 39, 36, 63, 75, 24].”
Taking Advantage of Multitask Learning for Fair Classification (2018, background): “In many recent papers [1–4 8–10 12 16 17 19 20 22–26 28 30 31 35–39] it has been shown how to enforce EOp⋄ constraints for ⋄∈{−,+}, during the learning phase of the model f ∈F .”
A Reductions Approach to Fair Classification (2018, neutral): “Dwork et al. (2018) also begin with a similar goal to ours.”
Empirical Risk Minimization under Fairness Constraints (2018, background): “Furthermore, the sensitive information could be correlated to the other features, hence removing the sensitive feature is not a guarantee for the fairness of the generated model (Dwork et al., 2018).”
Algorithmic Fairness (2020, methodology): “There are several previous papers in the domain of fair image classification [25, 46, 114, 124].”
REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasets (2020, background): “There has been signiﬁcant work looking at how to deal with this from the algorithm side (Dwork et al., 2012; Dwork et al., 2017; Khosla et al., 2012; Wang et al., 2020) with varying deﬁnitions of fairness (Gajane and Pech-enizkiy, 2017; Hardt et al., 2016; Kilbertus et al., 2017; Pleiss et al.,…”
Fairness risk measures (2019, methodology): “Several other recent works implicitly define perfect fairness in terms of subgroup risks (Dwork et al., 2018; Hashimoto et al., 2018; Alabi et al., 2018).”
InclusiveFaceNet: Improving Face Attribute Detection with Race and Gender Diversity (2017, methodology): “As in (Dwork et al., 2018), we use transfer learning.”
Repairing without Retraining: Avoiding Disparate Impact with Counterfactual Distributions (2019, methodology): “…between groups (see e.g., Lipton et al., 2018; Chouldechova, 2017; Kleinberg et al., 2016; Pleiss et al., 2017), and provides a suﬃcient testable condition that can indicate when to train diﬀerent classiﬁers for diﬀerent groups (see e.g., the methods of Dwork et al., 2018; Zafar et al., 2017b).”
Noise-tolerant fair classification (2019, background): “, 2019), and on means of achieving them (Zemel et al., 2013; Zafar et al., 2017b; Calmon et al., 2017; Dwork et al., 2018; Agarwal et al., 2018; Donini et al., 2018; Heidari et al., 2018).”
When optimizing nonlinear objectives is no harder than linear objectives (2018, methodology): “Despite such results, we know that we can design multi-objective metrics that trade-off fairness for accuracy and vice-versa (Woodworth et al. (2017); Menon and Williamson (2017); Dwork et al. (2018)).”
Fairness-Aware Programming (2019, methodology): “, by incorporating fairness into the objective function [10], eliminating proxies to sensitive attributes by preprocessing the data [11, 25], or applying a post-processing step to make a classifier fair [13].”
Trustworthy Artificial Intelligence: A Review (2022, methodology): “[58] proposed a decoupling technique that uses different classifiers for different groups to mitigate bias.”
Avoiding Resentment Via Monotonic Fairness (2019, background): “Similarly, [17] consider decoupled classifiers for separate classes, and how they can be combined to produce fair classification.”
Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (2019, methodology): “Conversely, [12] propose a scheme for decoupling classifiers, which we use to create our domain independent architecture.”
Making Decisions that Reduce Discriminatory Impacts (2019, unknown): “…there has been much recent work aimed at quantifying and removing discrimination (Berk et al., 2017; Bolukbasi et al., 2016; Chouldechova, 2017; Dwork et al., 2012; 2018; Edwards & Storkey, 2015; Hardt et al., 2016; Kamiran & Calders, 2009; Kamishima et al., 2012; Kilbertus et al., 2017;…”