A Convex Framework for Fair Regression
Authors:
Richard Berk,
Hoda Heidari,
Shahin Jabbari,
Matthew Joseph,
Michael Kearns,
Jamie Morgenstern,
Seth Neel,
Aaron Roth
Abstract:
We introduce a flexible family of fairness regularizers for (linear and logistic) regression problems. These regularizers all enjoy convexity, permitting fast optimization, and they span the rang from notions of group fairness to strong individual fairness. By varying the weight on the fairness regularizer, we can compute the efficient frontier of the accuracy-fairness trade-off on any given datas…
▽ More
We introduce a flexible family of fairness regularizers for (linear and logistic) regression problems. These regularizers all enjoy convexity, permitting fast optimization, and they span the rang from notions of group fairness to strong individual fairness. By varying the weight on the fairness regularizer, we can compute the efficient frontier of the accuracy-fairness trade-off on any given dataset, and we measure the severity of this trade-off via a numerical quantity we call the Price of Fairness (PoF). The centerpiece of our results is an extensive comparative study of the PoF across six different datasets in which fairness is a primary consideration.
△ Less
Submitted 7 June, 2017;
originally announced June 2017.
