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

arXiv cs.LG / 4/7/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper presents a reproducibility study that unifies multiple research lines on spurious correlations, shortcut learning, the Clever Hans effect, and group-distributional non-robustness to improve deep neural network reliability in high-stakes domains.
It compares recent correction methods—especially those using explainable AI (XAI)—against non-XAI baselines under difficult conditions such as limited data and severe subgroup imbalance.
The study finds that XAI-based approaches generally outperform non-XAI methods, with Counterfactual Knowledge Distillation (CFKD) performing most consistently for improved generalization across experiments.
Practical deployment is constrained because many methods depend on access to group labels, which are often infeasible to obtain manually and can be difficult for automated label discovery tools like Spectral Relevance Analysis (SpRAy) to handle under complex features and heavy imbalance.
The authors show that minority scarcity in validation sets makes model selection and hyperparameter tuning unreliable, highlighting a key barrier to deploying robust, trustworthy models in safety-critical applications.

Abstract

Deep Neural Networks (DNNs) are increasingly utilized in high-stakes domains like medical diagnostics and autonomous driving where model reliability is critical. However, the research landscape for ensuring this reliability is terminologically fractured across communities that pursue the same goal of ensuring models rely on causally relevant features rather than confounding signals. While frameworks such as distributionally robust optimization (DRO), invariant risk minimization (IRM), shortcut learning, simplicity bias, and the Clever Hans effect all address model failure due to spurious correlations, researchers typically only reference work within their own domains. This reproducibility study unifies these perspectives through a comparative analysis of correction methods under challenging constraints like limited data availability and severe subgroup imbalance. We evaluate recently proposed correction methods based on explainable artificial intelligence (XAI) techniques alongside popular non-XAI baselines using both synthetic and real-world datasets. Findings show that XAI-based methods generally outperform non-XAI approaches, with Counterfactual Knowledge Distillation (CFKD) proving most consistently effective at improving generalization. Our experiments also reveal that the practical application of many methods is hindered by a dependency on group labels, as manual annotation is often infeasible and automated tools like Spectral Relevance Analysis (SpRAy) struggle with complex features and severe imbalance. Furthermore, the scarcity of minority group samples in validation sets renders model selection and hyperparameter tuning unreliable, posing a significant obstacle to the deployment of robust and trustworthy models in safety-critical areas.