Research map / Architectures and efficiency / Pruning and efficient tuning
Exploring Nonlinear Pathway in Parameter Space for Machine Unlearning
Yingdan Shi, Ren Wang
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 MCUβ pathway search on top of NegGrad+ | NegGrad+ · CIFAR-10 10% random forgetting (PreResNet-110) | Avg. Gap: 2.97 -> 1.00 | Improved |
| add MCUβ on top of FT | FT · CIFAR-10 10% random forgetting | Avg. Gap: 6.57 -> 3.37 | Improved |
| add MCUβ on top of GA | GA · CIFAR-10 10% random forgetting | Avg. Gap: 7.31 -> 3.45 | Improved |
| add MCUβ on top of RL | RL · CIFAR-10 10% random forgetting | Avg. Gap: 2.70 -> 1.46 | Improved |
| add MCUβ on top of SalUn | SalUn · CIFAR-10 10% random forgetting | Avg. Gap: 5.99 -> 3.68 | Improved |
| adaptive beta instead of fixed beta (ablation) | MCU · CIFAR-10 10% random forgetting | Avg. Gap: 1.18 -> 1.00 | Improved |
| Bezier curve instead of Polychain pathway (ablation) | MCUβ · CIFAR-10 10% random forgetting | Avg. Gap: 1.28 -> 1.00 (Polychain-3 vs Bezier) | Improved |
| more Polychain control points (ablation) | MCUβ Polychain · CIFAR-10 10% random forgetting | Avg. Gap: 1.03 (c=1), 1.53 (c=2), 1.28 (c=3) | No clear effect |
| linear interpolation between original and pre-unlearning model instead of Bezier (ablation) | MCU · CIFAR-10 10% random forgetting | accuracy gap | Hurt |
| entire-parameter mask (10%) instead of no mask (ablation) | MCU · CIFAR-10 10% random forgetting | epoch time: 75.23% speedup | Improved |
| parameter mask instead of random 10% mask (ablation) | MCU · CIFAR-10 10% random forgetting | Dr accuracy drop at t=0.5: 3.44% drop with random mask | Improved |
| increase beta (ablation) | MCU · CIFAR-10 10% random forgetting | avg accuracy gap / effective region: gap 1.20 (0.1) -> 0.66 (0.3) | Mixed |
Closest papers by meaning
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.90)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · similarity 0.89)
- Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement (NeurIPS 2024 · similarity 0.89)
- The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning Evaluation (NeurIPS 2026 · similarity 0.88)
- Fast Machine Unlearning without Retraining through Selective Synaptic Dampening (AAAI 2024 · similarity 0.88)
- Machine Unlearning under Overparameterization (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.
