EUVIMEDEUVIMED entdecken
Crossref

Plug‐and‐Play Retain Selection for Large Language Model Unlearning: An Entanglement‐Driven Framework

Yunxuan Xiong, Hai Zhang

Expert Systems · 2026-08 · 43 (8)

DOI: 10.1111/exsy.70364 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Expert Systems, online erschienen am 12.07.2026. Verfasst von Yunxuan Xiong und 1 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine unlearning for large language models (LLMs) aims to selectively remove target knowledge while preserving non‐target knowledge. Existing methods often overlook the semantic entanglement between knowledge to be forgotten and knowledge to be retained, which may degrade retention performance. Motivated by this observation, retain‐subset construction is reformulated as an entanglement‐aware knowledge selection problem, and a plug‐and‐play framework is proposed. Specifically, the proposed framework quantifies knowledge coupling through a bounded Entanglement Score, and combines retrieval‐based selection with diversity‐aware sampling to construct representative retain subsets that require stronger protection. The selected data can be directly incorporated into existing unlearning algorithms, such as NPO and SimNPO, without modifying their loss functions. Experiments on TOFU and MUSE demonstrate improved utility–forgetting trade‐offs and reduced privacy leakage, while results on WMDP indicate that the framework is particularly beneficial when the knowledge to be forgotten and the knowledge to be retained are strongly coupled. Overall, this framework provides a general and plug‐and‐play approach to enhancing knowledge preservation in LLM unlearning.

Zitieren

APA

Xiong, Y., & Zhang, H. (2026). Plug‐and‐Play Retain Selection for Large Language Model Unlearning: An Entanglement‐Driven Framework. Expert Systems, 43(8). https://doi.org/10.1111/exsy.70364

Vancouver

Xiong Y, Zhang H. Plug‐and‐Play Retain Selection for Large Language Model Unlearning: An Entanglement‐Driven Framework. Expert Systems. 2026;43(8). doi:10.1111/exsy.70364

BibTeX
@article{Xiong2026,
  author = {Xiong, Yunxuan and Zhang, Hai},
  title = {Plug‐and‐Play Retain Selection for Large Language Model Unlearning: An Entanglement‐Driven Framework},
  journal = {Expert Systems},
  year = {2026},
  volume = {43},
  number = {8},
  doi = {10.1111/exsy.70364}
}
Kostenlos starten →