Research map / Architectures and efficiency / Pruning and efficient tuning
Efficient Knowledge Deletion from Trained Models Through Layer-wise Partial Machine Unlearning
Vinay Chakravarthi Gogineni, Esmaeil S. Nadimi
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 layer-wise pruning of stored updates (partial amnesiac unlearning) | Amnesiac unlearning, MLP · MNIST (unlearn class 3) | accuracy on retained data (%): 51.93 -> 94.47 | Improved |
| add layer-wise pruning of stored updates (partial amnesiac unlearning) | Amnesiac unlearning, ResNet9 · MNIST (unlearn class 3) | accuracy on retained data (%): 68.04 -> 98.33 | Improved |
| add layer-wise pruning of stored updates (partial amnesiac unlearning) | Amnesiac unlearning, AlexNet · OrganAMNIST (unlearn class 3) | accuracy on retained data (%): 9.90 -> 80.44 | Improved |
| add layer-wise pruning of stored updates (partial amnesiac unlearning) | Amnesiac unlearning, ConvNet · OrganAMNIST (unlearn class 3) | accuracy on retained data (%): 16.66 -> 75.20 | Improved |
| add layer-wise partial updates to label-flipping unlearning | Label-flipping unlearning, LeNet · MNIST (unlearn class 3) | accuracy on retained data (%): 86.61 -> 90.62 | Improved |
| add layer-wise partial updates to label-flipping unlearning | Label-flipping unlearning, ResNet9 · OrganAMNIST (unlearn class 3) | accuracy on retained data (%): 76.93 -> 82.27 | Improved |
| add layer-wise partial updates to optimization-based unlearning | Optimization-based unlearning, FCNN · MNIST (unlearn class 3) | accuracy on retained data (%): 85.12 -> 91.36 | Improved |
| add layer-wise partial updates to optimization-based unlearning | Optimization-based unlearning, ResNet9 · OrganAMNIST (unlearn class 3) | accuracy on retained data (%): 76.93 -> 85.89 | Improved |
| add layer-wise partial updates to optimization-based unlearning | Optimization-based unlearning, ResNet9 · MNIST (unlearn class 3) | accuracy on retained data (%): 97.33 -> 97.93 | No clear effect |
| add layer-wise partial updates to label-flipping unlearning | Label-flipping unlearning, ResNet9 · MNIST (unlearn class 3) | accuracy on retained data (%): 97.28 -> 97.88 | No clear effect |
| add layer-wise partial updates to label-flipping unlearning | Label-flipping unlearning, AlexNet · OrganAMNIST (unlearn class 3) | accuracy on retained data (%): 56.64 -> 56.33 | Mixed |
| use random layer-wise pruning with decreasing amount over depth instead of magnitude-based or global pruning (ablation) | Partial amnesiac unlearning · MNIST and OrganAMNIST | accuracy on retained data (%) | Improved |
Closest papers by meaning
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.92)
- Unlearning or Obfuscating? Jogging the Memory of Unlearned LLMs via Benign Relearning (ICLR 2025 · similarity 0.91)
- What makes unlearning hard and what to do about it (NeurIPS 2024 · similarity 0.91)
- Towards Scalable Exact Machine Unlearning Using Parameter-Efficient Fine-Tuning (ICLR 2025 · similarity 0.90)
- Label-Agnostic Forgetting: A Supervision-Free Unlearning in Deep Models (ICLR 2024 · similarity 0.90)
- Learning What to Forget: Improving LLM Unlearning via Learned Token-Level Importance (NeurIPS 2026 · 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.
