Research map / Continual learning / Class-incremental
Decoupling the Class Label and the Target Concept in Machine Unlearning
Jianing Zhu, Bo Han, Jiangchao Yao, Jianliang Xu, Gang Niu, Masashi Sugiyama
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 |
|---|---|---|---|
| use TARF instead of gradient ascent (GA) | ResNet-18 · CIFAR-10 target mismatch forgetting | average gap: 20.80 -> 1.23 | Improved |
| use TARF instead of fine-tuning (FT) | ResNet-18 · CIFAR-100 target mismatch forgetting | average gap: 28.54 -> 0.21 | Improved |
| use TARF instead of fine-tuning (FT) | ResNet-18 · CIFAR-10 model mismatch forgetting | average gap: 5.33 -> 2.90 | Improved |
| use TARF instead of FT | ResNet-18 · CIFAR-100 data mismatch forgetting | average gap: 37.15 -> 1.17 | Improved |
| use TARF instead of FT | ResNet-18 · CIFAR-10 all matched forgetting | average gap: 1.07 -> 1.01 | No clear effect |
| use TARF instead of SCRUB | ResNet-18 · CIFAR-100 all matched forgetting | average gap: 0.71 -> 1.11 | Hurt |
| use TARF instead of GA | ResNet-18 · CIFAR-100 data mismatch forgetting | average gap: 2.43 -> 1.17 | Improved |
| use TARF instead of GA | ResNet-18 · run-time efficiency CIFAR-10 all matched | RTE: 0.25 -> 4.21 | Hurt |
| use annealed (learning-rate-reduced) gradient ascent instead of constant k (ablation) | TARF · model mismatch forgetting | closeness to Retrained on forgetting data | Improved |
| use gradient cleaning (0) instead of gradient ascent on selected forgetting data (ablation) | TARF · target mismatch forgetting | retaining accuracy | Improved |
| increase initial ascent strength k beyond about 0.05 (ablation) | TARF · CIFAR-100 all matched forgetting | gap | Hurt |
Closest papers by meaning
- Forgetting Has Neighbors: Localized Collateral Forgetting in Machine Unlearning (NeurIPS 2026 · similarity 0.91)
- Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance (NeurIPS 2026 · similarity 0.91)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.90)
- Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks (CVPR 2025 · similarity 0.90)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.89)
- Fast Machine Unlearning without Retraining through Selective Synaptic Dampening (AAAI 2024 · similarity 0.89)
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.
