Superclass-Guided Representation Disentanglement for Spurious Correlation Mitigation

1. 1Identify superclasses that lie higher in the semantic hierarchy than the task's actual labels
2. 2Incorporate superclasses into the model to guide representation disentanglement
3. 3Use the disentangled representations to mitigate spurious correlations and improve group robustness
4. 4Evaluate the model's performance on test domains with different group distributions

ML researchers and engineers on a team benefit from this approach as it enhances group robustness to spurious correlations, and can be applied to real-world problems where group annotations are not available or reliable

💡 Leveraging superclasses can be a more effective way to mitigate spurious correlations than relying on auxiliary group annotations

💡 Mitigate spurious correlations with Superclass-Guided Representation Disentanglement!