Research map / Continual learning / Continual fine-tuning of LLMs
Can Bad Teaching Induce Forgetting? Unlearning in Deep Networks Using an Incompetent Teacher
Read the paperCheck 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 |
|---|---|---|---|
| Use competent and incompetent teachers to selectively transfer retain and forget information | ResNet18 unlearning · CIFARSuper20, Rocket forgotten | D_f accuracy: 82 -> 2±0.40 | Helped |
| Use competent and incompetent teachers to selectively transfer retain and forget information | ResNet18 unlearning · CIFARSuper20, Rocket forgotten | D_r accuracy: 85.78 -> 85.05±0.61 | No clear effect |
| Use competent and incompetent teachers to selectively transfer retain and forget information | ResNet18 unlearning · CIFARSuper20, Rocket forgotten | ZRF: 0.87 -> 0.99 | Helped |
| Replace ResNet34 incompetent teacher with ResNet18 (ablation) | Competent/incompetent teacher unlearning · CIFARSuper20, Rocket forgotten | D_f accuracy: 1 -> 15 | Mixed |
| Replace ResNet34 incompetent teacher with a random predictor (ablation) | Competent/incompetent teacher unlearning · CIFARSuper20, Rocket forgotten | D_f accuracy: 1 -> 5 | Mixed |
| Replace a randomly initialized teacher with a partially trained teacher (ablation) | Competent/incompetent teacher unlearning · CIFARSuper20, Rocket forgotten | JS-Divergence: 0.04 -> 0.02 | Helped |
| Increase retain data used from 50% to 100% (ablation) | LSTM unlearning · Human Activity Recognition, Person 1 forgotten, 1 epoch | ZRF: 0.69 -> 0.79 | Helped |
| Increase unlearning epochs from 1 to 5 (ablation) | LSTM unlearning · Human Activity Recognition, Person 1 forgotten, 10% retain data | ZRF: 0.62 -> 0.75 | Helped |
| Increase learning rate from 0.0001 to 0.1 (ablation) | LSTM unlearning · Human Activity Recognition, Person 1 forgotten, 2 epochs and 30% retain data | ZRF: 0.59 -> 0.96 | Helped |
| Use a partially trained incompetent teacher instead of a randomly initialized teacher (ablation) | Class-level unlearning · CIFAR10, one class forgotten | D_f accuracy: 4.22 -> 0 | Helped |
Closest papers by meaning
- What makes unlearning hard and what to do about it (NeurIPS 2024 · similarity 0.89)
- Forgetting Has Neighbors: Localized Collateral Forgetting in Machine Unlearning (NeurIPS 2026 · similarity 0.89)
- When to Forget? Complexity Trade-offs in Machine Unlearning (ICML 2025 · similarity 0.89)
- Decoupled Distillation to Erase: A General Unlearning Method for Any Class-centric Tasks (CVPR 2025 · similarity 0.89)
- Not All Wrong is Bad: Using Adversarial Examples for Unlearning (ICML 2025 · similarity 0.88)
- From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization (NeurIPS 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.