Spurious Correlation-Aware Embedding Regularization for Worst-Group Robustness

Researchers propose SCER (Spurious Correlation-Aware Embedding Regularization), a new deep learning approach that improves AI model robustness by regularizing feature representations to suppress spurious correlations. The method demonstrates superior performance in worst-group accuracy across vision and language tasks compared to existing state-of-the-art approaches.

• →SCER addresses the vulnerability of deep learning models to distribution shifts by directly regularizing embedding representations.
• →The method provides a theoretical framework connecting embedding space representations with worst-group error performance.
• →SCER encourages models to focus on core features while reducing sensitivity to spurious patterns through embedding-level constraints.
• →Systematic evaluation shows SCER outperforms prior state-of-the-art methods in worst-group accuracy across multiple domains.
• →The approach is particularly effective for subpopulation shift scenarios where models struggle with underrepresented groups.