Multicalibrated Regression for Downstream Fairness

TL;DR

This paper presents a method to convert multicalibrated regression functions into classifiers that satisfy various fairness constraints efficiently, using minimal additional data and computation, and handling intersecting groups.

Contribution

It extends multicalibration to enable post-processing for fairness-constrained classification, including intersecting groups, without labeled data.

Findings

Post-processing achieves fairness constraints with minimal unlabeled data.

Method handles intersecting groups, generalizing prior work.

Computational complexity comparable to solving a single fair learning task.

Abstract

We show how to take a regression function $\hat{f}$ that is appropriately ``multicalibrated'' and efficiently post-process it into an approximately error minimizing classifier satisfying a large variety of fairness constraints. The post-processing requires no labeled data, and only a modest amount of unlabeled data and computation. The computational and sample complexity requirements of computing $\hat f$ are comparable to the requirements for solving a single fair learning task optimally, but it can in fact be used to solve many different downstream fairness-constrained learning problems efficiently. Our post-processing method easily handles intersecting groups, generalizing prior work on post-processing regression functions to satisfy fairness constraints that only applied to disjoint groups. Our work extends recent work showing that multicalibrated regression functions are ``omnipredictors'' (i.e. can be post-processed to optimally solve unconstrained ERM problems) to constrained optimization.