統一多模態模型的概念遺忘能否統一?探索跨路徑概念遺忘
Is Forgetting Unified? Exploring Concept Unlearning in Unified Multimodal Models
一項研究提出 UniForget,旨在統一多模態模型中同時抑制目標概念的生成與識別,並保留其他能力。研究亦推出 UUBench,涵蓋 30 個目標概念及 60 個語義相關概念;UniForget 在 BAGEL-7B-MoT 和 Janus-Pro-7B 上的聯合遺忘率分別為 80.0% 和 92.7%。分析顯示,聯合遺忘與能力保留之間存在取捨。
Published on Nov 3
Authors:
,
,
,
,
,
,
,
,
Abstract
Cross-pathway concept unlearning in unified multimodal models is addressed by a new benchmark and method that jointly suppresses target concept generation and recognition while preserving other capabilities.
Unified 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.
View arXiv page View PDF Add to collection
Models citing this paper 0
No model linking this paper
Cite arxiv.org/abs/2611.04467 in a model README.md to link it from this page.
Datasets citing this paper 0
No dataset linking this paper
Cite arxiv.org/abs/2611.04467 in a dataset README.md to link it from this page.
Spaces citing this paper 0
No Space linking this paper
Cite arxiv.org/abs/2611.04467 in a Space README.md to link it from this page.
Collections including this paper 0
No Collection including this paper
Add this paper to a collection to link it from this page.
來源:HuggingFace Daily Papers(社區熱門論文) · huggingface.co