Research map / Continual learning / Continual fine-tuning of LLMs
Towards Unbounded Machine Unlearning
Meghdad Kurmanji, Peter Triantafillou, Jamie Hayes, Eleni Triantafillou
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 rewinding to SCRUB for selective unlearning (ablation) | SCRUB · CIFAR-10 selective, ResNet | Basic MIA accuracy (%): 78.00 -> 58.67 | Improved |
| add rewinding to SCRUB for class unlearning (ablation) | SCRUB · CIFAR-10 class, ResNet | Basic MIA accuracy (%): 52.20 -> 52.20 | No clear effect |
| use SCRUB instead of Bad-T for resolving confusion | unlearning method · CIFAR-5 RC ResNet | Fgt test error (%): 27.0 -> 19.7 | Improved |
| use SCRUB instead of fine-tuning on retain set | unlearning method · CIFAR-5 RC ResNet | Fgt test error (%): 79.3 -> 19.7 | Improved |
| use contrastive teacher-student objective with alternating max/min steps instead of plain NegGrad+ | unlearning · CIFAR-5 RC ResNet | Fgt test error (%): 69.0 -> 19.7 | Improved |
| use Bad-T (move toward bad teacher) instead of SCRUB | unlearning · RC ResNet | IC test error (%): 19.0 -> 47.3 | Hurt |
| vary SCRUB hyperparameters alpha and gamma over [0.1,3] (ablation) | SCRUB · CIFAR-10 | retain error | No clear effect |
Closest papers by meaning
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.92)
- Towards Source-Free Machine Unlearning (CVPR 2025 · similarity 0.92)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.92)
- Adversarial Machine Unlearning (ICLR 2025 · similarity 0.91)
- The Unseen Threat: Residual Knowledge in Machine Unlearning under Perturbed Samples (NeurIPS 2025 · similarity 0.91)
- What makes unlearning hard and what to do about it (NeurIPS 2024 · 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.
