Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911203337437184 |
|---|---|
| author | Wang, Changsheng Zhang, Yihua Jia, Jinghan Ram, Parikshit Wei, Dennis Yao, Yuguang Pal, Soumyadeep Baracaldo, Nathalie Liu, Sijia |
| author_facet | Wang, Changsheng Zhang, Yihua Jia, Jinghan Ram, Parikshit Wei, Dennis Yao, Yuguang Pal, Soumyadeep Baracaldo, Nathalie Liu, Sijia |
| contents | Machine unlearning offers a promising solution to privacy and safety concerns in large language models (LLMs) by selectively removing targeted knowledge while preserving utility. However, current methods are highly sensitive to downstream fine-tuning, which can quickly recover forgotten information-even from unrelated tasks. To address this, we introduce invariance into unlearning for the first time, inspired by invariant risk minimization (IRM). Building on this principle, we propose invariant LLM unlearning (ILU), a regularization-based framework that enhances robustness. Notably, ILU generalizes well to diverse fine-tuning tasks, even when trained using a single dataset. A task vector analysis is also provided to further elucidate the rationale behind ILU's effectiveness. Extensive experiments on the WMDP and MUSE benchmark, reveal that ILU significantly outperforms state-of-the-art unlearning methods, including negative preference optimization (NPO) and representation misdirection for unlearning (RMU). Notably, ILU achieves superior unlearning robustness across diverse downstream fine-tuning scenarios (e.g., math, paraphrase detection, and sentiment analysis) while preserving the fine-tuning performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_01339 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning Wang, Changsheng Zhang, Yihua Jia, Jinghan Ram, Parikshit Wei, Dennis Yao, Yuguang Pal, Soumyadeep Baracaldo, Nathalie Liu, Sijia Machine Learning Machine unlearning offers a promising solution to privacy and safety concerns in large language models (LLMs) by selectively removing targeted knowledge while preserving utility. However, current methods are highly sensitive to downstream fine-tuning, which can quickly recover forgotten information-even from unrelated tasks. To address this, we introduce invariance into unlearning for the first time, inspired by invariant risk minimization (IRM). Building on this principle, we propose invariant LLM unlearning (ILU), a regularization-based framework that enhances robustness. Notably, ILU generalizes well to diverse fine-tuning tasks, even when trained using a single dataset. A task vector analysis is also provided to further elucidate the rationale behind ILU's effectiveness. Extensive experiments on the WMDP and MUSE benchmark, reveal that ILU significantly outperforms state-of-the-art unlearning methods, including negative preference optimization (NPO) and representation misdirection for unlearning (RMU). Notably, ILU achieves superior unlearning robustness across diverse downstream fine-tuning scenarios (e.g., math, paraphrase detection, and sentiment analysis) while preserving the fine-tuning performance. |
| title | Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2506.01339 |