Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.09172 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915783680983040 |
|---|---|
| author | Shao, Pengyang Zhai, Naixin Chen, Lei Yang, Yonghui Zhu, Fengbin Yang, Xun Wang, Meng |
| author_facet | Shao, Pengyang Zhai, Naixin Chen, Lei Yang, Yonghui Zhu, Fengbin Yang, Xun Wang, Meng |
| contents | As Large Language Models (LLMs) increasingly shape online content, removing targeted information from well-trained LLMs (also known as LLM unlearning) has become critical for web governance. A key challenge lies in sample-wise imbalance within the forget set: different samples exhibit widely varying unlearning difficulty, leading to asynchronous forgetting where some knowledge remains insufficiently erased while others become over-forgotten. To address this, we propose BalDRO, a novel and efficient framework for balanced LLM unlearning. BalDRO formulates unlearning as a min-sup process: an inner step identifies a worst-case data distribution that emphasizes hard-to-unlearn samples, while an outer step updates model parameters under this distribution. We instantiate BalDRO via two efficient variants: BalDRO-G, a discrete GroupDRO-based approximation focusing on high-loss subsets, and BalDRO-DV, a continuous Donsker-Varadhan dual method enabling smooth adaptive weighting within standard training pipelines. Experiments on TOFU and MUSE show that BalDRO significantly improves both forgetting quality and model utility over existing methods, and we release code for reproducibility. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_09172 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning Shao, Pengyang Zhai, Naixin Chen, Lei Yang, Yonghui Zhu, Fengbin Yang, Xun Wang, Meng Machine Learning As Large Language Models (LLMs) increasingly shape online content, removing targeted information from well-trained LLMs (also known as LLM unlearning) has become critical for web governance. A key challenge lies in sample-wise imbalance within the forget set: different samples exhibit widely varying unlearning difficulty, leading to asynchronous forgetting where some knowledge remains insufficiently erased while others become over-forgotten. To address this, we propose BalDRO, a novel and efficient framework for balanced LLM unlearning. BalDRO formulates unlearning as a min-sup process: an inner step identifies a worst-case data distribution that emphasizes hard-to-unlearn samples, while an outer step updates model parameters under this distribution. We instantiate BalDRO via two efficient variants: BalDRO-G, a discrete GroupDRO-based approximation focusing on high-loss subsets, and BalDRO-DV, a continuous Donsker-Varadhan dual method enabling smooth adaptive weighting within standard training pipelines. Experiments on TOFU and MUSE show that BalDRO significantly improves both forgetting quality and model utility over existing methods, and we release code for reproducibility. |
| title | BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2601.09172 |