Average Individual Fairness: Algorithms, Generalization and Experiments

TL;DR

This paper introduces a new fairness framework for classification that considers distributions over individuals and tasks, providing algorithms with theoretical guarantees and empirical validation for fair learning.

Contribution

It proposes a novel fairness definition combining statistical and individual notions, along with an oracle-efficient algorithm and generalization guarantees for classification.

Findings

Algorithm effectively enforces fairness constraints.

Theoretical guarantees for generalization to new individuals and tasks.

Empirical results confirm the algorithm's effectiveness.

Abstract

We propose a new family of fairness definitions for classification problems that combine some of the best properties of both statistical and individual notions of fairness. We posit not only a distribution over individuals, but also a distribution over (or collection of) classification tasks. We then ask that standard statistics (such as error or false positive/negative rates) be (approximately) equalized across individuals, where the rate is defined as an expectation over the classification tasks. Because we are no longer averaging over coarse groups (such as race or gender), this is a semantically meaningful individual-level constraint. Given a sample of individuals and classification problems, we design an oracle-efficient algorithm (i.e. one that is given access to any standard, fairness-free learning heuristic) for the fair empirical risk minimization task. We also show that given sufficiently many samples, the ERM solution generalizes in two directions: both to new individuals, and to new classification tasks, drawn from their corresponding distributions. Finally we implement our algorithm and empirically verify its effectiveness.