Research map / Label-efficient and robust learning / Recognition and long-tail
ERM-KTP: Knowledge-Level Machine Unlearning via Knowledge Transfer
Shen Lin, Xiaoyu Zhang, Chenyang Chen, Xiaofeng Chen, Willy Susilo
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 the ERM structure to the CNN | CNN classifiers · CIFAR10, CIFAR100, and Tiny-ImageNet validation | accuracy: 89.6% -> 92.6%; 76.2% -> 77.6%; 49.3% -> 53.4% | Improved |
| Use ERM-KTP instead of retraining from scratch | ResNet-20 · CIFAR10, 40% classes removed | unlearning time: 668s -> 124s | Improved |
| Use ERM-KTP for class-specific unlearning | ResNet-20 · CIFAR10, 40% classes removed | target-class accuracy: 0.0% | Improved |
| Use ERM-KTP for class-specific unlearning | ResNeXt-50 · CIFAR100, 40% classes removed | target-class accuracy: 0.0% | Improved |
| Use ERM-KTP for class-specific unlearning | ResNet-50 · Tiny-ImageNet, 40% classes removed | target-class accuracy: 0.0% | Improved |
| Use ERM-KTP instead of Fine-tune | ResNeXt-50 · CIFAR100, 40% classes removed | validation accuracy: 83.4% -> 79.4% | Hurt |
| Use ERM-KTP instead of RfS | ResNet-50 · Tiny-ImageNet, 40% classes removed | validation accuracy: 45.1% -> 53.8% | Improved |
| Use ERM-KTP instead of RfS | ResNet-20 · CIFAR10, 40% classes removed | validation accuracy: 96.3% -> 96.1% | No clear effect |
| Use both L1-norm and inner-product regularization terms (ablation) | ERM-CNN · CIFAR100 | CSI | Improved |
| Remove the inner-product regularization term (ablation) | ERM-CNN · CIFAR100 | CSI | Hurt |
| Remove the L1-norm regularization term (ablation) | ERM-CNN · CIFAR100 | CSI | Hurt |
| Replace ERM-KTP with Fine-tune or random labels | unlearning methods · CIFAR10, ResNet-20 relearning experiment | relearning time: about the 50th epoch; around the 100th epoch | Hurt |
Closest papers by meaning
- Layer Attack Unlearning: Fast and Accurate Machine Unlearning via Layer Level Attack and Knowledge Distillation (AAAI 2024 · similarity 0.88)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.88)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.87)
- Efficient Source-free Unlearning via Energy-Guided Data Synthesis and Discrimination-Aware Multitask Optimization (ICML 2025 · similarity 0.87)
- The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples (NeurIPS 2025 · similarity 0.87)
- Towards Source-Free Machine Unlearning (CVPR 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.
