Rethinking LLM Unlearning: From Safety Constraints to Functional Utility Xiaoyu Xu, Minxin Du, Li Bai, Junxu Liu, Huadi Zheng, Peizhao Hu, Qingqing Ye, Haibo Hu
(Survey) (Paper)
My first survey on LLM unlearning, and the first to systematically cover functional forgetting: We organize prior work into two categories based on application focus: safety forgetting, including privacy erasure, copyright removal, and harmful suppression, and functional forgetting, including behavior probing, graceful forgetting, and model regularization. We analyze the first category along three dimensions, covering benchmark, unlearning algorithm, and evaluation; and the second category through three representative functional objectives: CoT faithfulness, learning with forgetting, and model regularizer.
When Routine Chats Turn Toxic: Unintended Long-Term State Poisoning in Personalized Agents Xiaoyu Xu, Minxin Du, Qipeng Xie, Haobin Ke, Qingqing Ye, Haibo Hu (arXiv 2026) (Paper)
FIT to Forget: Robust Continual Unlearning for Large Language Models Xiaoyu Xu, Minxin Du, Kun Fang, Yaxin Xiao, Zhicong Huang, Cheng Hong, Qingqing Ye, Haibo Hu (arXiv 2026) (Paper)
From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning Xiaoyu Xu, Minxin Du, Zitong Li, Zi Liang, Zhibiao Guo, Shiyu Zhang, Peizhao Hu, Qingqing Ye, Haibo Hu (ACL 2026 Findings) (Paper)
Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMs Xiaoyu Xu, Xiang Yue, Yang Liu, Qingqing Ye, Huadi Zheng, Peizhao Hu, Minxin Du, Haibo Hu (ICML 2026) (Paper)
OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models Xiaoyu Xu, Minxin Du, Qingqing Ye, Haibo Hu (EMNLP 2025) (Paper)
Coauthor
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning Maggie Huan, Yuetai Li, Tuney Zheng, Xiaoyu Xu, Seungone Kim, Minxin Du, Radha Poovendran, Graham Neubig, Xiang Yue (ICML 2026) (Paper)
