Research map / Continual learning / Class-incremental
Can Vision Models Truly Forget? Mirage: Representation-Level Certification of Visual Unlearning
Zhenyu Yu, yangchen zeng, Chunlei Meng, Guangzhen Yao, Shuigeng Zhou
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 |
|---|---|---|---|
| run BU for more unlearning epochs (1 to 20) (ablation) | BU · COVID-19 | LPR gap vs Retrain: +15.9 -> +15.9 (LPR stays above 94%) | No clear effect |
| run BU for more unlearning epochs (1 to 20) (ablation) | BU · CIFAR-10 | retained accuracy Accr: 79.2 -> 21.8 | Hurt |
| run BU for more unlearning epochs (1 to 20) (ablation) | BU · CIFAR-10 | LPR gap vs Retrain: +4.4 -> +0.6 | Mixed |
| increase number of passive parties K from 2 to 8 (ablation) | BU · CIFAR-10 | LPR gap vs Retrain: +4.8 -> +3.7 | No clear effect |
| increase number of passive parties K from 2 to 8 (ablation) | Target · CIFAR-10 | LPR gap vs Retrain: +1.2 -> -11.9 | Hurt |
| increase number of passive parties K from 2 to 8 (ablation) | FT · CIFAR-10 | LPR gap vs Retrain: +7.9 -> +6.2 | No clear effect |
| apply BU instead of retraining | BU · COVID-19 | LPR: 79.2 -> 94.7 | Hurt |
| apply FT or SSD unlearning | SSD · COVID-19 | feature separability: .031 -> .340 | Hurt |
| apply Target unlearning | Target · COVID-19 | LPR: 79.2 -> 69.9 | Mixed |
| forget class-level vs sample-level | all methods · 7 datasets | LPR: 99% on MNIST, above 96% on COVID-19 vs about 50% | Mixed |
Closest papers by meaning
- Targeted Forgetting of Image Subgroups in CLIP Models (CVPR 2025 · similarity 0.89)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.89)
- The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation (NeurIPS 2026 · similarity 0.87)
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 2025 · similarity 0.87)
- Measuring Representational Shifts in Continual Learning: A Linear Transformation Perspective (ICML 2025 · similarity 0.87)
- The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples (NeurIPS 2025 · similarity 0.87)
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.
