الملخص
Concept erasure aims to suppress undesirable knowledge in text-to-image generative models. However, existing robustness evaluations typically rely on relearning attacks tailored to specific model architectures. We study concept reactivation across two substantially different generative paradigms: noise-prediction U-Nets and flow-matching Transformers. We introduce \textbf{Concept Score Relearning (CSR)}, a unified parameter-level framework that reactivates erased concepts by optimizing each model within its native prediction space. CSR requires no external target-concept image dataset and applies the same concept-directed objective to both U-Net-based Stable Diffusion and Transformer-based FLUX. Experiments across diverse concepts and multiple erasure methods demonstrate consistent concept reactivation across both architectures, highlighting the cross-architecture applicability of CSR and the persistent recoverability of apparently erased concepts. For strict nudity, CSR reaches average ASRs of 50.47\% on FLUX and 40.29\% on Stable Diffusion, consistently ranking first across all evaluated safety settings.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Tae, H. X., Zhang, J., He, W., Wang, X., & Lim, W. Y. B. (2026). Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure. https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure
MLA 9
Tae, Hong Xi, et al. "Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure." https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure.
شيكاغو (المؤلف–التاريخ)
Tae, Hong Xi, Jiaming Zhang, Wenwen He, Xuan Wang, and Wei Yang Bryan Lim. 2026. "Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure." https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure.
هارفارد
Tae, H. X., Zhang, J., He, W., Wang, X. and Lim, W. Y. B. (2026) 'Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure', Available at: https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure.
فانكوفر
Tae HX, Zhang J, He W, Wang X, Lim WYB. Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure. https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure
IEEE
H. X. Tae, J. Zhang, W. He, X. Wang, and W. Y. B. Lim, "Concept Score Relearning: A Unified Cross-Architecture Attack on Concept Erasure," https://omanscience.com/ar/articles/concept-score-relearning-a-unified-cross-architecture-attack-on-concept-erasure.