Research map / Optimization / Implicit bias and generalization
Eternal Sunshine of the Spotless Net: Selective Forgetting in Deep Networks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Fine-tune on the retain set instead of using the proposed scrubbing methods | Original model · CIFAR-10 and Lacuna-10 | error on the forget and retain sets | Hurt |
| Use negative-gradient updates on the forget set | Original model · CIFAR-10 and Lacuna-10 | re-learn time | Hurt |
| Assign random labels to the forget-set samples and fine-tune | Original model · CIFAR-10 and Lacuna-10 | re-learn time | Hurt |
| Assign random labels to the forget-set samples and fine-tune | Original model · CIFAR-10 and Lacuna-10 | confidence entropy on forget samples | Hurt |
| Use Fisher-based noise scrubbing | Original model · CIFAR-10 and Lacuna-10 | re-learn time | Helped |
| Use variational noise scrubbing | Original model · CIFAR-10 and Lacuna-10 | re-learn time | Helped |
| Increase the forgetting parameter λ (ablation) | Fisher or variational scrubbing · ResNet; forgetting class 5 | information remaining about the forget cohort | Helped |
| Increase the forgetting parameter λ (ablation) | Fisher or variational scrubbing · ResNet; forgetting class 5 | error on retain samples | Hurt |
| Forget a larger cohort (ablation) | Fisher or variational scrubbing · CIFAR-10 and Lacuna-10 | information remaining about the forget cohort | Hurt |
| Pre-train before fine-tuning and scrubbing (ablation) | Deep network scrubbing · Local forgetting bound | information bound | Helped |
| Use a diagonal Fisher approximation for a full Newton step (ablation) | Fisher forgetting · Deep networks | approximation quality for the Newton step | Hurt |
| Use a diagonal Fisher approximation for adding noise (ablation) | Fisher forgetting · Deep networks | noise approximation quality | Helped |
Closest papers by meaning
- Mixed-Privacy Forgetting in Deep Networks (CVPR 2021 · similarity 0.91)
- Certified Unlearning for Neural Networks (ICML 2025 · similarity 0.86)
- Fast Machine Unlearning without Retraining through Selective Synaptic Dampening (AAAI 2024 · similarity 0.85)
- The Utility and Complexity of In- and Out-of-Distribution Machine Unlearning (ICLR 2025 · similarity 0.85)
- Distribution-Level Feature Distancing for Machine Unlearning: Towards a Better Trade-off Between Model Utility and Forgetting (AAAI 2025 · similarity 0.84)
- On Local Overfitting and Forgetting in Deep Neural Networks (AAAI 2025 · similarity 0.84)
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.