From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Xu, Xiaoyu, Du, Minxin, Li, Zitong, Liang, Zi, Guo, Zhibiao, Zhang, Shiyu, Hu, Peizhao, Ye, Qingqing, Hu, Haibo
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917419406065664
author Xu, Xiaoyu
Du, Minxin
Li, Zitong
Liang, Zi
Guo, Zhibiao
Zhang, Shiyu
Hu, Peizhao
Ye, Qingqing
Hu, Haibo
author_facet Xu, Xiaoyu
Du, Minxin
Li, Zitong
Liang, Zi
Guo, Zhibiao
Zhang, Shiyu
Hu, Peizhao
Ye, Qingqing
Hu, Haibo
contents Although machine unlearning is essential for removing private, harmful, or copyrighted content from LLMs, current benchmarks often fail to faithfully represent the true ``forgetting scope'' learned by the model. We formalize two distinct unlearning granularities, domain-level and instance-level, and propose \BiForget, an automated framework for synthesizing high-quality forget sets. Unlike prior work relying on \emph{external} generators, \BiForget exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting. Our experiments across diverse benchmarks show that it achieves a superior balance of relevance, diversity, and efficiency. Quantitatively, in the Harry Potter domain, it improves relevance by ${\sim}20$ and diversity by ${\sim}$0.05 while \emph{halving} the total data size compared to SOTAs. Ultimately, it facilitates more robust forgetting and better utility preservation, providing a more rigorous foundation for evaluating LLM unlearning.
format Preprint
id arxiv_https___arxiv_org_abs_2601_04278
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning
Xu, Xiaoyu
Du, Minxin
Li, Zitong
Liang, Zi
Guo, Zhibiao
Zhang, Shiyu
Hu, Peizhao
Ye, Qingqing
Hu, Haibo
Computation and Language
Artificial Intelligence
Cryptography and Security
Machine Learning
Although machine unlearning is essential for removing private, harmful, or copyrighted content from LLMs, current benchmarks often fail to faithfully represent the true ``forgetting scope'' learned by the model. We formalize two distinct unlearning granularities, domain-level and instance-level, and propose \BiForget, an automated framework for synthesizing high-quality forget sets. Unlike prior work relying on \emph{external} generators, \BiForget exploits the target model per se to elicit data that matches its internal knowledge distribution through seed-guided and adversarial prompting. Our experiments across diverse benchmarks show that it achieves a superior balance of relevance, diversity, and efficiency. Quantitatively, in the Harry Potter domain, it improves relevance by ${\sim}20$ and diversity by ${\sim}$0.05 while \emph{halving} the total data size compared to SOTAs. Ultimately, it facilitates more robust forgetting and better utility preservation, providing a more rigorous foundation for evaluating LLM unlearning.
title From Domains to Instances: Dual-Granularity Data Synthesis for LLM Unlearning
topic Computation and Language
Artificial Intelligence
Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2601.04278