Primary
Erased but Recoverable: On the Fragility of Concept Erasure Under Fine-Tuning
Primary source. Published by the vendor, CERT, regulator, government agency or researcher responsible.
arXiv:2611.24220v1 Announce Type: new Abstract: Existing concept erasure methods for text-to-image diffusion models primarily focus on suppressing target concepts while preserving non-target generation quality. However, in practical applications, open-source diffusion models are often further adapted through DreamBooth, LoRA, or Custom Diffusion. Such adaptations can reshape internal…
Read the full story at arXiv cs.CR (Cryptography & Security) ↗HackerNews24 shows a short summary only. The full article, its copyright and any corrections belong to the original publisher.