OxEnsemble: Fair Ensembles for Low-Data Classification

TL;DR

OxEnsemble is a novel, data-efficient ensemble method that enforces fairness in low-data, unbalanced classification tasks, especially in medical imaging, with theoretical guarantees and improved fairness-accuracy trade-offs.

Contribution

The paper introduces OxEnsemble, a new ensemble approach that efficiently enforces fairness in low-data regimes with theoretical validation and superior empirical performance.

Findings

OxEnsemble achieves more consistent fairness outcomes.

It provides stronger fairness-accuracy trade-offs.

The method is both data- and compute-efficient.

Abstract

We address the problem of fair classification in settings where data is scarce and unbalanced across demographic groups. Such low-data regimes are common in domains like medical imaging, where false negatives can have fatal consequences. We propose a novel approach \emph{OxEnsemble} for efficiently training ensembles and enforcing fairness in these low-data regimes. Unlike other approaches, we aggregate predictions across ensemble members, each trained to satisfy fairness constraints. By construction, \emph{OxEnsemble} is both data-efficient -- carefully reusing held-out data to enforce fairness reliably -- and compute-efficient, requiring little more compute than used to fine-tune or evaluate an existing model. We validate this approach with new theoretical guarantees. Experimentally, our approach yields more consistent outcomes and stronger fairness-accuracy trade-offs than existing methods across multiple challenging medical imaging classification datasets.