titleIs Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models
authorsZhongkai Xue, Junkai Chen, Mingcong Lei, Rumo Zhang, Yixiao Liu, Zhimin Sun, Shu Wu, Qiang Liu, Lihua Zhang, Liang Wang
date2026-11-03
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.04467
affiliationChinese Academic of Science Institute of Automation
upvotes2 (Hugging Face Daily Papers, 2026-11-05)
linksabs pdf hf
cite
abstractUnified multimodal models (UMMs) integrate image generation and visual understanding, but are existing single-pathway unlearning methods sufficient to suppress target concepts across both? In this work, we study cross-pathway concept unlearning. Our goal is to prevent a model from both generating images of a target concept and identifying it in input images, while preserving other capabilities. To this end, we introduce UUBench, a diagnostic benchmark covering 30 target concepts and 60 semantically related neighboring concepts. Experiments on representative UMMs show that the evaluated single-pathway methods achieve limited joint forgetting: generation-side editors strongly suppress target generation but yield limited forgetting in visual understanding, while understanding-side baselines struggle to achieve substantial forgetting even in their own pathway. Their direct composition also yields limited joint forgetting. We therefore propose UniForget, achieving joint forgetting rates of 80.0% on BAGEL-7B-MoT and 92.7% on Janus-Pro-7B, outperforming representative baselines. Further analysis reveals trade-offs between joint forgetting and capability preservation.

bibtex

@misc{xue2026is,
  title={Is Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models},
  author={Zhongkai Xue and Junkai Chen and Mingcong Lei and Rumo Zhang and Yixiao Liu and Zhimin Sun and Shu Wu and Qiang Liu and Lihua Zhang and Liang Wang},
  year={2026},
  eprint={2611.04467},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.04467}
}