Research map / Optimization / Implicit bias and generalization
Gradient sparsity regularization for training unlearning-compatible models
Sobhan Hemati, Soufiane Lamghari, Masoud Asgharian, Xu Li, Hongliang Li
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 |
|---|---|---|---|
| Train with gradient l1 regularization before unlearning | FT unlearning · CIFAR-10 class forgetting (ResNet-18) | Avg. Gap: 22.61 -> 0.81 | Improved |
| Train with gradient l1 regularization before unlearning | IU unlearning · CIFAR-10 class forgetting (ResNet-18) | Avg. Gap: 40.61 -> 5.76 | Improved |
| Train with gradient l1 regularization before unlearning | WS unlearning · CIFAR-10 random forgetting (ResNet-18) | Avg. Gap: 7.87 -> 2.06 | Improved |
| Train with gradient l1 regularization before unlearning | FF unlearning · CIFAR-10 random forgetting (ResNet-18) | Avg. Gap: 10.83 -> 10.59 | No clear effect |
| Train with gradient l1 regularization before unlearning | FT unlearning · CIFAR-10 random forgetting (VGG16) | Avg. Gap: 3.54 -> 4.14 | Hurt |
| Train with gradient l1 regularization before unlearning | GA, IU, FF, FT · CIFAR-100 random forgetting (ResNet-18) | Avg. Gap: FT 15.08 -> 14.34 | No clear effect |
| Train with gradient l1 regularization before unlearning | GA unlearning · ImageNet class forgetting (ResNet-18) | Avg. Gap: 9.53 -> 5.72 | Improved |
| Train with gradient l1 regularization before unlearning | FT unlearning · ImageNet class forgetting (ResNet-18) | Avg. Gap: 17.20 -> 16.81 | No clear effect |
| Train with gradient l1 regularization before unlearning | IU unlearning · SVHN class forgetting (ViT-small) | Avg. Gap: 42.36 -> 29.1 | Improved |
| Add gradient sparsity regularization under backdoor poisoning | MobileNetV2 · Poisoned CIFAR-10 (3%) | poison test accuracy: 86.62 -> 24.63 | Improved |
| Add gradient sparsity regularization under backdoor poisoning | MobileNetV2 · Poisoned CIFAR-10 (3%) | clean test accuracy: 68.19 -> 68.1 | No clear effect |
| Use gradient l1 norm instead of l2 norm regularizer (ablation) | FT unlearning · CIFAR-10 class forgetting | Avg. Gap | Improved |
Closest papers by meaning
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.92)
- Towards LLM Unlearning Resilient to Relearning Attacks: A Sharpness-Aware Minimization Perspective and Beyond (ICML 2025 · similarity 0.90)
- Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement (NeurIPS 2024 · similarity 0.89)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.88)
- Hessian-Free Online Certified Unlearning (ICLR 2025 · similarity 0.88)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · similarity 0.88)
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.
