Research map / Optimization / Optimizers
Machine Unlearning via Adaptive Gradient Reweighting and Multi-stage Objective Optimization
Juxin Lu, Haoyu Shi, Mengyao Wang, Huaiwen Zhang
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 Adaptive Gradient Reweighting (ablation) | unlearning with multi-stage optimization · CIFAR-10 10% random forgetting | Avg Gap: 2.78 -> 1.57 | Improved |
| add Adaptive Gradient Reweighting (ablation) | unlearning with multi-stage optimization · CIFAR-10 single-class forgetting | Gap Score: 0.205 -> 0.020 | Improved |
| add Direction Rectification (ablation) | unlearning method · CIFAR-10 10% random forgetting | Avg Gap: 3.37 -> 1.57 | Improved |
| add Direction Rectification (ablation) | unlearning method · CIFAR-10 single-class forgetting | Gap Score: 0.420 -> 0.020 | Improved |
| add Temporal Stabilization (ablation) | unlearning method · CIFAR-10 10% random forgetting | Avg Gap: 3.49 -> 1.57 | Improved |
| add Temporal Stabilization (ablation) | unlearning method · CIFAR-10 single-class forgetting | Gap Score: 0.118 -> 0.020 | Improved |
| add Adaptive Objective Combination (ablation) | unlearning method · CIFAR-10 10% random forgetting | Avg Gap: 3.13 -> 1.57 | Improved |
| add Adaptive Objective Combination (ablation) | unlearning method · CIFAR-10 single-class forgetting | Gap Score: 0.740 -> 0.020 | Improved |
| add all components (ablation) | baseline unlearning · CIFAR-10 single-class forgetting | RAL: -3.14 -> +0.28 | Improved |
| scale number of forgotten classes from 1 to 20 | Ours · CIFAR-100 | H-Mean: 80.47 (1 class) -> 74.72 (20 classes) | No clear effect |
| increase forget ratio to 50% | baselines vs Ours · CIFAR-100 (ResNet18) | Avg Gap: LoTUS 17.25 vs Ours 8.88 | Improved |
| apply to medical unlearning | Ours vs original · ISIC 2019 | Mean confidence: 97.1 -> 89.1 | Improved |
Closest papers by meaning
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.92)
- Efficient Utility-Preserving Machine Unlearning with Implicit Gradient Surgery (NeurIPS 2025 · similarity 0.92)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.91)
- Fast Machine Unlearning without Retraining through Selective Synaptic Dampening (AAAI 2024 · similarity 0.90)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · similarity 0.90)
- Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement (NeurIPS 2024 · similarity 0.90)
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.
