Optimized Data Pre-Processing for Discrimination Prevention

By Flávio du Pin Calmon, Dennis Wei, Karthikeyan Natesan Ramamurthy, Kush R. Varshney. 2017

64 citations recorded by the source corpus.

Non-discrimination is a recognized objective in algorithmic decision making. In this paper, we introduce a novel probabilistic formulation of data pre-processing for reducing discrimination. We propose a convex optimization for learning a data transformation with three goals: controlling discrimination, limiting distortion in individual data samples, and preserving utility. We characterize the impact of limited sample size in accomplishing this objective, and apply two instances of the proposed optimization to datasets, including one on real-world criminal recidivism. The results demonstrate that all three criteria can be simultaneously achieved and also reveal interesting patterns of bias in American society.

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

Learning Adversarially Fair and Transferable Representations (2018, methodology): “Dwork et al. [2012] notably discussed individual fairness within the owner-vendor framework we utilize. Zemel et al. [2013] proposed encouraging elements of both group and individual fairness via a regularized training procedure.”
Fairness in Criminal Justice Risk Assessments: The State of the Art (2017, background): “…of dependence is subject to two constraints: (1) the joint distribution of the transformed variables is very close to the joint distribution of the original predictors and (2) no individual cases are substantially distorted because large changes are made in predictor values (Calmon et al., 2017).”
Learning Controllable Fair Representations (2018, background): “Calmon et al. (2017) transform the dataset to prevent discrimination on specific classification tasks.”
Active Fairness in Algorithmic Decision Making (2018, background): “Consequently, substantial work on defining and optimizing for algorithmic fairness has surged in recent years, where most methods fall in one of three broad categories: a) adjustments by post-processing the output of black-box models [12], b) introducing constraints during the learning process [2, 2”
Conditional Contrastive Learning: Removing Undesirable Information in Self-Supervised Representations (2021, background): “While most of the study focuses on a supervised setup [11, 16, 27], this paper studies the self-supervised setup where downstream tasks are unknown, but the learned representations still need to preserve fairness criterion [4, 26].”
Provably Fair Representations (2017, methodology): “It may be appropriate to require group fairness only on some subset of the input data [3], such as requiring equal employment outcomes for men and women with suitable qualifications rather than for all men and women.”