Research map / Adversarial robustness and security / Attacks on text and graphs
Unlearning’s Blind Spots: Over‑Unlearning and Prototypical Relearning Attack
SeungBum Ha, Saerom Park, Sung Whan Yoon
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 Spotter regularizers to SalUn | SalUn · CIFAR-10 (one class unlearned) | OU@epsilon: 0.1664 -> 0.0345 | Improved |
| add Spotter regularizers to DELETE | DELETE · CIFAR-10 (one class unlearned) | OU@epsilon: 0.1216 -> 0.0232 | Improved |
| add Spotter regularizers to UNSC | UNSC · CIFAR-10 (one class unlearned) | OU@epsilon: 0.1575 -> 0.0089 | Improved |
| add Spotter regularizers to SalUn | SalUn · CIFAR-10 (one class unlearned) | Proto-Acc_f: 95.34 -> 0.34 | Improved |
| add Spotter regularizers to UNSC | UNSC · CIFAR-100 (ten classes unlearned) | Proto-Acc_f: 73.62 -> 18.54 | Improved |
| add over-unlearning loss L_o (lambda1 from 1 to 0.7) to SalUn + Spotter (ablation) | SalUn + Spotter · CIFAR-10 (one class unlearned), lambda2=0 | OU@epsilon: 0.1864 -> 0.0335 | Improved |
| add L_o (lambda1 from 1 to 0.3) to Spotter with lambda2=1 (ablation) | Spotter · CIFAR-10 (one class unlearned) | OU@epsilon: 0.0485 -> 0.0197 | Improved |
| add dispersion loss L_sim (lambda2 from 0 to 1) to Spotter lambda1=0.7 (ablation) | Spotter · CIFAR-10 (one class unlearned) | Proto-Acc_f: 99.70 -> 0.26 | Improved |
| add dispersion loss L_sim (lambda2 from 0 to 1) to Spotter lambda1=0.7 (ablation) | Spotter · CIFAR-10 (one class unlearned) | Acc_rt: 94.19 -> 93.91 | Hurt |
| increase lambda2 from 1 to 2 for Spotter lambda1=1 (ablation) | Spotter · CIFAR-10 (one class unlearned) | Acc_rt: 93.71 -> 91.93 | Hurt |
| apply Spotter to pretrained CLIP ViT-B/32 | CLIP ViT-B/32 · ImageNet class unlearning | Acc_ft: 54.18 -> 0.00 | Improved |
| add Spotter to SalUn on retained test accuracy | SalUn · CIFAR-10 (one class unlearned) | Acc_rt: 89.09 -> 93.37 | Improved |
Closest papers by meaning
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 2025 · similarity 0.93)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.92)
- Unlearning or Obfuscating? Jogging the Memory of Unlearned LLMs via Benign Relearning (ICLR 2025 · similarity 0.91)
- The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples (NeurIPS 2025 · similarity 0.91)
- Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond (ICML 2025 · similarity 0.91)
- Keeping an Eye on LLM Unlearning: The Hidden Risk and Remedy (NeurIPS 2025 · similarity 0.91)
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.
