Decoupled classifiers for fair and efficient machine learning

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

44 citations recorded by the source corpus.

When it is ethical and legal to use a sensitive attribute (such as gender or race) in machine learning systems, the question remains how to do so. We show that the naive application of machine learning algorithms using sensitive features leads to an inherent tradeoff in accuracy between groups. We provide a simple and efficient decoupling technique, that can be added on top of any black-box machine learning algorithm, to learn different classifiers for different groups. Transfer learning is used to mitigate the problem of having too little data on any one group. The method can apply to a range of fairness criteria. In particular, we require the application designer to specify as joint loss function that makes explicit the trade-off between fairness and accuracy. Our reduction is shown to efficiently find the minimum loss as long as the objective has a certain natural monotonicity property which may be of independent interest in the study of fairness in algorithms.

Subfields: Artificial Intelligence, Safety Research

How researchers see this paper

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

In Artificial Intelligence

In Safety Research

Citations in the corpus

From Parity to Preference-based Notions of Fairness in Classification (2017, background): “Additionally, preferred treatment allows group-conditional decision making (not allowed by treatment parity), which is necessary to achieve high decision accuracy in scenarios when the predictive power of features varies greatly between different sensitive user groups [13], as shown in Figure 1.”
Multiaccuracy: Black-Box Post-Processing for Fairness in Classification (2018, background): “These works fit into the larger literature on fairness in learning and prediction tasks [2, 7, 8, 11, 19, 26, 33], discussions of the utility-fairness tradeoffs in fair classification [1, 3–5, 20, 25].”
Predict Responsibly: Increasing Fairness by Learning To Defer (2017, background): “Several works have focused on the goal of treating similar people similarly (individual fairness) and the resulting necessity of fair-awareness (Dwork et al., 2011; Zemel et al., 2013; Dwork et al., 2017).”
Does mitigating ML's impact disparity require treatment disparity? (2017, background): “[11] address the problem of how best to incorporate information about protected characteristics for several of these other fairness criteria.”
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 [17,18,36,62] with varying definitions of fairness [21,24,37,50,68] that are often deemed to be mathematically incompatible with each other [14, 38], but in this work, we look at the problem earlier in the pipel”
Fair Forests: Regularized Tree Induction to Minimize Model Bias (2017, result): “This makes our results comparable with a larger body of work, as their approach and metrics have been widely used through the literature [1, 6, 10, 19].”
A Framework for Understanding Unintended Consequences of Machine Learning (2019, methodology): “…learning meth-ods , such as multitask learning, that parameterize different groups differently in the model deﬁnition and facilitate learning multiple simpler functions that take into account group differences (Gelman, 2006; Dwork et al., 2017; Suresh, Gong, and Guttag, 2018; Oneto et al., 2018).”
Algorithm fairness in artificial intelligence for medicine and healthcare (2023, background): “And bias-mitigation strategies that consider the inclusion of protected attributes are few 94,168–171 .”