Mitigating Spurious Correlations in Machine Learning Models: Techniques and Applications
Mashrin Srivastava · 2023
Spurious correlations in machine learning models can lead to undesirable behaviors and catastrophic failures in real-world applications. This paper aims to provide an overview of techniques and methodologies for mitigating spurious correlations in machine learning models, with a focus on invariance and stability. We discuss the challenges and limitations of existing approaches and present case studies from various domains, including medical imaging, visual question answering, and natural language processing. Finally, we propose future research directions to address the issue of spurious correlations in machine learning models.