Research map / Adversarial robustness and security / Attacks on text and graphs
ROVER: Robust Generative Continual Identity Unlearning Against Relearning Attacks
Tairan Huang, Qiang Chen, Beibei Hu, Yunlong Zhao, Hongyan Xu, Zhiyuan Chen, Yi Chen, Xiu Su
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Similarity Navigator routing between O-Res and PGN (ablation) | ROVER · CelebAHQ ablation | ID similarity: 0.27 -> 0.05 | Improved |
| add Phantom Guard Network (ablation) | ROVER · CelebAHQ ablation | ID similarity: 0.21 -> 0.05 | Improved |
| add Residual Orthogonal Resonator (ablation) | ROVER · CelebAHQ ablation | ID similarity: 0.16 -> 0.05 | Improved |
| replace GUIDE with ROVER under relearning attack | GUIDE · CelebAHQ relearning attack | ID similarity: 0.89 -> 0.05 | Improved |
| replace GUIDE with ROVER in benign continual unlearning | GUIDE · CelebAHQ benign continual unlearning | FID_real: 4.93 -> 3.42 | Improved |
| increase number of continual unlearning requests | GUIDE · CelebAHQ | ID similarity: 0.17 -> 0.23 | Hurt |
| increase number of continual unlearning requests | ROVER · CelebAHQ | ID similarity: 0.09 -> 0.15 | Hurt |
| set similarity threshold epsilon to 0.3 or 0.7 instead of 0.5 (ablation) | Similarity Navigator · CelebAHQ | ID similarity: 0.05 -> 0.25 (eps 0.3), 0.21 (eps 0.7) | Hurt |
| set orthogonality weight to values other than 0.01 (ablation) | O-Res · CelebAHQ | ID similarity: 0.05 (0.01) vs 0.09 (0.001), 0.21 (1.0) | Mixed |
Closest papers by meaning
- Generative Unlearning for Any Identity (CVPR 2024 · similarity 0.88)
- Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond (ICML 2025 · similarity 0.88)
- Adversarial Mixup Unlearning (ICLR 2025 · similarity 0.86)
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 2025 · similarity 0.86)
- Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models (NeurIPS 2024 · similarity 0.86)
- Unlearn and Burn: Adversarial Machine Unlearning Requests Destroy Model Accuracy (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
