A Novel Regularization Approach to Fair ML

TL;DR

This paper presents a simple, flexible regularization method called Explicitly Deweighted Features (EDF) for fair machine learning, allowing customizable fairness-utility tradeoffs across various algorithms.

Contribution

It introduces EDF, a novel, easy-to-understand regularization technique that adjusts feature weights to promote fairness, adaptable to multiple ML models.

Findings

EDF effectively reduces bias in models.

The method allows customizable fairness-utility tradeoffs.

A new criterion for evaluating fairness protection is proposed.

Abstract

A number of methods have been introduced for the fair ML issue, most of them complex and many of them very specific to the underlying ML moethodology. Here we introduce a new approach that is simple, easily explained, and potentially applicable to a number of standard ML algorithms. Explicitly Deweighted Features (EDF) reduces the impact of each feature among the proxies of sensitive variables, allowing a different amount of deweighting applied to each such feature. The user specifies the deweighting hyperparameters, to achieve a given point in the Utility/Fairness tradeoff spectrum. We also introduce a new, simple criterion for evaluating the degree of protection afforded by any fair ML method.