1 paper
Amy Lu, Changxiu Ji
Association unlearning aims to disable learned label-attribute shortcuts while preserving task performance. Existing evaluations mainly measure output-level robustness or probe whe…