HSFM: Hard-Set-Guided Feature-Space Meta-Learning for Robust Classification under Spurious Correlations

Deep neural networks often rely on spurious features to make predictions, which makes them brittle under distribution shift and on samples where the spurious correlation does not hold (e.g., minority-group examples). Recent studies have shown that, even in such settings, the feature extractor of an Empirical Risk Minimization (ERM)-trained model can learn rich and informative representations, and that much of the failure may be attributed to the classifier head. In particular, retraining a lightweight head while keeping the backbone frozen can substantially improve performance on shifted distributions and minority groups. Motivated by this observation, we propose a bilevel meta-learning method that performs augmentation directly in feature space to improve spurious correlation handling in the classifier head. Our method learns support-side feature edits such that, after a small number of inner-loop updates on the edited features, the classifier achieves lower loss on hard examples and improved worst-group performance. By operating at the backbone output rather than in pixel space or through end-to-end optimization, the method is highly efficient and stable, requiring only a few minutes of training on a single GPU. We further validate our method with CLIP-based visualizations, showing that the learned feature-space updates induce semantically meaningful shifts aligned with spurious attributes.

Analyzing Effects of Mixed Sample Data Augmentation on Model Interpretability

Data augmentation strategies are actively used when training deep neural networks (DNNs). Recent studies suggest that they are effective at various tasks. However, the effect of data augmentation on DNNs' interpretability is not yet widely investigated. In this paper, we explore the relationship between interpretability and data augmentation strategy in which models are trained with different data augmentation methods and are evaluated in terms of interpretability. To quantify the interpretability, we devise three evaluation methods based on alignment with humans, faithfulness to the model, and the number of human-recognizable concepts in the model. Comprehensive experiments show that models trained with mixed sample data augmentation show lower interpretability, especially for CutMix and SaliencyMix augmentations. This new finding suggests that it is important to carefully adopt mixed sample data augmentation due to the impact on model interpretability, especially in mission-critical applications.