Shortcut Mitigation via Spurious-Positive Samples

TL;DR

This paper introduces a targeted analysis method to identify and mitigate reliance on spurious features in models, improving robustness without needing balanced data or annotations.

Contribution

It proposes a novel approach to detect and regularize neurons associated with spurious features, reducing reliance on misleading cues during training.

Findings

Improved model robustness against spurious correlations.

No need for group-balanced held-out data or annotations.

Effective identification of neurons related to spurious attributes.

Abstract

Shortcut mitigation strategies commonly rely on training data annotations, group-balanced held-out data or the presence of all groups, i.e., all combinations of (spurious) attributes and classes, in the training data. However, these requirements are rarely met in practice. We instead propose a method for targeted model analysis to identify a small set of instances in which the model relies on spurious attributes. Using that set and following ``this feature should not be used for prediction'' reasoning, we identify highly relevant neurons in an intermediate layer and regularize their impact. This ensures that models learn to depend on informative features rather than being right for the wrong reasons, thereby improving robustness without requiring additional balanced held-out data or annotations.