FairSAM: Fair Classification on Corrupted Data Through Sharpness-Aware Minimization

TL;DR

FairSAM tackles the dual challenges of robustness and fairness in corrupted image classification. It introduces Corrupted Degradation Disparity to quantify subgroup differences in degradation and integrates fairness into Sharpness-Aware Minimization via instance-level reweighting, producing equitable robustness across demographic groups. The method demonstrates superior fairness-robustness trade-offs on CelebA, FairFace, and LFW under multiple noise types and levels, including out-of-distribution transfers. This work provides a principled framework for achieving both high accuracy and fair subgroup performance when deployment conditions involve corrupted data, with potential impact on real-world ethical AI systems.

Abstract

Image classification models trained on clean data often suffer from significant performance degradation when exposed to testing corrupted data, such as images with impulse noise, Gaussian noise, or environmental noise. This degradation not only impacts overall performance but also disproportionately affects various demographic subgroups, raising critical algorithmic bias concerns. Although robust learning algorithms like Sharpness-Aware Minimization (SAM) have shown promise in improving overall model robustness and generalization, they fall short in addressing the biased performance degradation across demographic subgroups. Existing fairness-aware machine learning methods - such as fairness constraints and reweighing strategies - aim to reduce performance disparities but hardly maintain robust and equitable accuracy across demographic subgroups when faced with data corruption. This reveals an inherent tension between robustness and fairness when dealing with corrupted data. To address these challenges, we introduce one novel metric specifically designed to assess performance degradation across subgroups under data corruption. Additionally, we propose \textbf{FairSAM}, a new framework that integrates \underline{Fair}ness-oriented strategies into \underline{SAM} to deliver equalized performance across demographic groups under corrupted conditions. Our experiments on multiple real-world datasets and various predictive tasks show that FairSAM successfully reconciles robustness and fairness, offering a structured solution for equitable and resilient image classification in the presence of data corruption.

Figures (2)

• Figure 1: Comparison of corrupted degradation and accuracy between subgroups using various methods.
• Figure 2: Comparison of SAM-based methods among varying noise levels. FairSAM achieves comparable or even better accuracy while maintaining the lowest bias $\Delta p$.

Theorems & Definitions (1)

• Definition 1: Corrupted Degradation Disparity