Probing Unlearned Diffusion Models: A Transferable Adversarial Attack Perspective

arXiv:2404.19382v1 Announce Type: new
Abstract: Advanced text-to-image diffusion models raise safety concerns regarding identity privacy violation, copyright infringement, and Not Safe For Work content generation. Towards this, unlearning methods have been developed to erase these involved concepts from diffusion models. However, these unlearning methods only shift the text-to-image mapping and preserve the visual content within the generative space of diffusion models, leaving a fatal flaw for restoring these erased concepts. This erasure trustworthiness problem needs probe, but previous methods are …

abstract advanced adversarial arxiv concepts concerns content generation copyright copyright infringement cs.cv diffusion diffusion models however identity image image diffusion infringement mapping perspective privacy raise safe safety shift text text-to-image type unlearning work