Reproducibility study on how to find Spurious Correlations, Shortcut Learning, Clever Hans or Group-Distributional non-robustness and how to fix them

A reproducibility study unifies research on spurious correlations in deep neural networks across different domains, comparing correction methods including XAI-based approaches. The research finds that Counterfactual Knowledge Distillation (CFKD) most effectively improves model generalization, though practical deployment remains challenging due to group labeling dependencies and data scarcity issues.

• →XAI-based correction methods generally outperform non-XAI approaches for addressing spurious correlations in neural networks.
• →Counterfactual Knowledge Distillation (CFKD) proved most consistently effective at improving model generalization across datasets.
• →Many correction methods are hindered by dependency on group labels, as manual annotation is often infeasible in practice.
• →Automated tools like Spectral Relevance Analysis struggle with complex features and severe data imbalance.
• →Scarcity of minority group samples makes model selection and hyperparameter tuning unreliable for safety-critical applications.