defense arXiv Jan 28, 2026 · 9w ago
Lulu Xue, Shengshan Hu, Wei Lu et al. · Huazhong University of Science and Technology · Institute of Guizhou Aerospace Measuring and Testing Technology +2 more
Defends machine unlearning against inversion attacks that reconstruct erased training data via cosine-space perturbations
Model Inversion Attack vision
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.
cnn transformer Huazhong University of Science and Technology · Institute of Guizhou Aerospace Measuring and Testing Technology · University of Technology Sydney +1 more
