Label Noise Robustness for Domain-Agnostic Fair Corrections via Nearest Neighbors Label Spreading

TL;DR

This paper introduces a label noise correction method using label spreading on nearest neighbors, significantly improving worst-group accuracy in last-layer retraining for fairness across noisy labels and diverse datasets.

Contribution

It proposes a novel label spreading correction technique for last-layer retraining that is robust to label noise and computationally efficient.

Findings

Achieves state-of-the-art worst-group accuracy under symmetric label noise.

Effective across various datasets with spurious correlations.

Minimal additional computational overhead.

Abstract

Last-layer retraining methods have emerged as an efficient framework for correcting existing base models. Within this framework, several methods have been proposed to deal with correcting models for subgroup fairness with and without group membership information. Importantly, prior work has demonstrated that many methods are susceptible to noisy labels. To this end, we propose a drop-in correction for label noise in last-layer retraining, and demonstrate that it achieves state-of-the-art worst-group accuracy for a broad range of symmetric label noise and across a wide variety of datasets exhibiting spurious correlations. Our proposed approach uses label spreading on a latent nearest neighbors graph and has minimal computational overhead compared to existing methods.