Research map / Architectures and efficiency / Pruning and efficient tuning
Model Sparsity Can Simplify Machine Unlearning
Jinghan Jia, Jiancheng Liu, Parikshit Ram, Yuguang Yao, Gaowen Liu, Yang Liu, Pranay Sharma, Sijia Liu
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 |
|---|---|---|---|
| Prune to 95% sparsity with OMP before fine-tuning | FT · CIFAR-10, ResNet-18, class-wise forgetting | Disparity Ave. ↓: 25.78 -> 11.50 | Improved |
| Prune to 95% sparsity with OMP before influence unlearning | IU · CIFAR-10, ResNet-18, class-wise forgetting | Disparity Ave. ↓: 5.41 -> 1.10 | Improved |
| Prune to 95% sparsity with OMP before fine-tuning | FT · CIFAR-10, ResNet-18, random data forgetting | Disparity Ave. ↓: 2.74 -> 1.92 | Improved |
| Prune to 95% sparsity with OMP before influence unlearning | IU · CIFAR-10, ResNet-18, random data forgetting | Disparity Ave. ↓: 4.07 -> 3.08 | Improved |
| Add constant ℓ1 sparsity regularization during fine-tuning (ablation) | MUSparse · CIFAR-10, ResNet-18, random data forgetting (10%) | UA: 6.60 (1.19) | Mixed |
| Increase ℓ1 regularization linearly during fine-tuning (ablation) | MUSparse · CIFAR-10, ResNet-18, random data forgetting (10%) | UA: 3.80 (1.61) | Mixed |
| Decrease ℓ1 regularization linearly during fine-tuning (ablation) | MUSparse · CIFAR-10, ResNet-18, random data forgetting (10%) | UA: 5.35 (0.06) | Improved |
| Use SynFlow pruning instead of OMP before fine-tuning (ablation) | FT · CIFAR-10, ResNet-18, class-wise forgetting | UA and MIAF | Mixed |
| Use IMP pruning instead of OMP before fine-tuning (ablation) | FT · CIFAR-10, ResNet-18, class-wise forgetting | UA and MIAF | Hurt |
| Prune to 95% sparsity before approximate unlearning | GA · CIFAR-10, ResNet-18, class-wise forgetting | Disparity Ave. ↓: 7.12 -> 6.41 | Improved |
| Prune to 95% sparsity before approximate unlearning | FF · CIFAR-10, ResNet-18, class-wise forgetting | Disparity Ave. ↓: 5.32 -> 1.98 | Improved |
| Prune to 95% sparsity before approximate unlearning | GA · CIFAR-10, ResNet-18, random data forgetting | Disparity Ave. ↓: 4.26 -> 3.05 | Improved |
Closest papers by meaning
- LoRA Unlearns More and Retains More (Student Abstract) (AAAI 2025 · similarity 0.92)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.91)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.91)
- Towards Scalable Exact Machine Unlearning Using Parameter-Efficient Fine-Tuning (ICLR 2025 · similarity 0.91)
- What makes unlearning hard and what to do about it (NeurIPS 2024 · similarity 0.91)
- LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertainty (CVPR 2025 · 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.
