Robust LLM Unlearning with MUDMAN: Meta-Unlearning with Disruption Masking And Normalization
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917111893327872 |
|---|---|
| author | Sondej, Filip Yang, Yushi Kniejski, Mikołaj Windys, Marcel |
| author_facet | Sondej, Filip Yang, Yushi Kniejski, Mikołaj Windys, Marcel |
| contents | Language models can retain dangerous knowledge and skills even after extensive safety fine-tuning, posing both misuse and misalignment risks. Recent studies show that even specialized unlearning methods can be easily reversed. To address this, we systematically evaluate many existing and novel components of unlearning methods and identify ones crucial for irreversible unlearning.
We introduce Disruption Masking, a technique in which we only allow updating weights, where the signs of the unlearning gradient and the retaining gradient are the same. This ensures all updates are non-disruptive.
Additionally, we identify the need for normalizing the unlearning gradients, and also confirm the usefulness of meta-learning. We combine these insights into MUDMAN (Meta-Unlearning with Disruption Masking and Normalization) and validate its effectiveness at preventing the recovery of dangerous capabilities. MUDMAN outperforms the prior TAR method by 40%, setting a new state-of-the-art for robust unlearning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_12484 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Robust LLM Unlearning with MUDMAN: Meta-Unlearning with Disruption Masking And Normalization Sondej, Filip Yang, Yushi Kniejski, Mikołaj Windys, Marcel Machine Learning Artificial Intelligence Computation and Language Language models can retain dangerous knowledge and skills even after extensive safety fine-tuning, posing both misuse and misalignment risks. Recent studies show that even specialized unlearning methods can be easily reversed. To address this, we systematically evaluate many existing and novel components of unlearning methods and identify ones crucial for irreversible unlearning. We introduce Disruption Masking, a technique in which we only allow updating weights, where the signs of the unlearning gradient and the retaining gradient are the same. This ensures all updates are non-disruptive. Additionally, we identify the need for normalizing the unlearning gradients, and also confirm the usefulness of meta-learning. We combine these insights into MUDMAN (Meta-Unlearning with Disruption Masking and Normalization) and validate its effectiveness at preventing the recovery of dangerous capabilities. MUDMAN outperforms the prior TAR method by 40%, setting a new state-of-the-art for robust unlearning. |
| title | Robust LLM Unlearning with MUDMAN: Meta-Unlearning with Disruption Masking And Normalization |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2506.12484 |