UnlearnShield: Shielding Forgotten Privacy against Unlearning Inversion
This addresses a critical privacy threat for users of machine unlearning systems, offering a dedicated defense against inversion attacks.
The paper tackles the privacy vulnerability in machine unlearning where adversaries can reconstruct erased data via unlearning inversion, and proposes UnlearnShield as a defense that reduces inversion risk while maintaining model accuracy and forgetting efficacy.
Machine unlearning is an emerging technique that aims to remove the influence of specific data from trained models, thereby enhancing privacy protection. However, recent research has uncovered critical privacy vulnerabilities, showing that adversaries can exploit unlearning inversion to reconstruct data that was intended to be erased. Despite the severity of this threat, dedicated defenses remain lacking. To address this gap, we propose UnlearnShield, the first defense specifically tailored to counter unlearning inversion. UnlearnShield introduces directional perturbations in the cosine representation space and regulates them through a constraint module to jointly preserve model accuracy and forgetting efficacy, thereby reducing inversion risk while maintaining utility. Experiments demonstrate that it achieves a good trade-off among privacy protection, accuracy, and forgetting.