Optimal Transport Reweighting for Robust Learning under Spurious Correlations and Label Noise

2026-10-01

AI Summary

This study addresses the degradation of model performance caused by spurious correlations and label noise under subgroup shifts. To tackle this, we propose POTER, a framework grounded in optimal transport theory. Rather than relying on loss signals susceptible to noise interference, POTER computes sample weights based on the geometric relationships of instance-level transport alignment to assess importance. Notably, it achieves robust learning through a single round of empirical risk minimization (ERM), departing from conventional retraining paradigms. Experimental results demonstrate that POTER attains state-of-the-art worst-group accuracy across standard benchmarks and noisy settings, effectively mitigating the adverse effects of label contamination in minority groups.

Abstract

Machine learning models often suffer performance degradation under subpopulation shift, particularly when spurious correlations cause models to rely on shortcut features that fail to generalize across subgroups. A recent line of work mitigates this issue by using loss-based signals to identify informative samples, but these signals can become severely distorted under label noise: mislabeled samples may also incur large losses and contaminate subsequent reweighting or retraining. Despite its practical importance, this intersection remains largely underexplored. We propose POTER, a reweighting framework based on optimal transport that derives sample importance from the transport geometry between the training distribution and a reference distribution constructed from limited validation group annotations. By measuring alignment at the individual-sample level rather than relying on loss, POTER downweights mislabeled or strongly bias-aligned samples while assigning higher importance to samples better aligned with the reference distribution. In addition, POTER requires only a single ERM training stage, moving beyond the retraining paradigm common in recent work. Across standard benchmarks and noisy-label settings, POTER achieves state-of-the-art worst-group accuracy, including cases where label corruption is concentrated within minority subgroups.

Problem

• Spurious Correlations
• Label Noise
• Subpopulation Shift
• Robust Learning
• Worst-group Accuracy

Innovation

• Optimal Transport
• Sample Reweighting
• Spurious Correlations
• Label Noise
• Worst-group Accuracy