Equal Opportunity in Online Classification with Partial Feedback

TL;DR

This paper introduces an online classification algorithm that ensures fairness constraints like equal opportunity are met at every step, despite only observing partial feedback, and analyzes the associated regret bounds.

Contribution

It presents the first analysis of fairness constraints in online classification with partial feedback, providing bounds and an efficient algorithm for this setting.

Findings

The algorithm achieves near-optimal regret bounds under fairness constraints.

Fairness constraints impose a mild additional cost in regret compared to unconstrained settings.

Theoretical bounds demonstrate the feasibility of fair online classification with partial feedback.

Abstract

We study an online classification problem with partial feedback in which individuals arrive one at a time from a fixed but unknown distribution, and must be classified as positive or negative. Our algorithm only observes the true label of an individual if they are given a positive classification. This setting captures many classification problems for which fairness is a concern: for example, in criminal recidivism prediction, recidivism is only observed if the inmate is released; in lending applications, loan repayment is only observed if the loan is granted. We require that our algorithms satisfy common statistical fairness constraints (such as equalizing false positive or negative rates -- introduced as "equal opportunity" in Hardt et al. (2016)) at every round, with respect to the underlying distribution. We give upper and lower bounds characterizing the cost of this constraint in terms of the regret rate (and show that it is mild), and give an oracle efficient algorithm that achieves the upper bound.