Rethinking the Vulnerability of Concept Erasure and a New Method
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912624484024320 |
|---|---|
| author | Richardson, Alex D. Zhang, Kaicheng Beerens, Lucas Chen, Dongdong |
| author_facet | Richardson, Alex D. Zhang, Kaicheng Beerens, Lucas Chen, Dongdong |
| contents | The proliferation of text-to-image diffusion models has raised significant privacy and security concerns, particularly regarding the generation of copyrighted or harmful images. In response, concept erasure (defense) methods have been developed to "unlearn" specific concepts through post-hoc finetuning. However, recent concept restoration (attack) methods have demonstrated that these supposedly erased concepts can be recovered using adversarially crafted prompts, revealing a critical vulnerability in current defense mechanisms. In this work, we first investigate the fundamental sources of adversarial vulnerability and reveal that vulnerabilities are pervasive in the prompt embedding space of concept-erased models, a characteristic inherited from the original pre-unlearned model. Furthermore, we introduce **RECORD**, a novel coordinate-descent-based restoration algorithm that consistently outperforms existing restoration methods by up to 17.8 times. We conduct extensive experiments to assess its compute-performance tradeoff and propose acceleration strategies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2502_17537 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Rethinking the Vulnerability of Concept Erasure and a New Method Richardson, Alex D. Zhang, Kaicheng Beerens, Lucas Chen, Dongdong Machine Learning Artificial Intelligence Cryptography and Security The proliferation of text-to-image diffusion models has raised significant privacy and security concerns, particularly regarding the generation of copyrighted or harmful images. In response, concept erasure (defense) methods have been developed to "unlearn" specific concepts through post-hoc finetuning. However, recent concept restoration (attack) methods have demonstrated that these supposedly erased concepts can be recovered using adversarially crafted prompts, revealing a critical vulnerability in current defense mechanisms. In this work, we first investigate the fundamental sources of adversarial vulnerability and reveal that vulnerabilities are pervasive in the prompt embedding space of concept-erased models, a characteristic inherited from the original pre-unlearned model. Furthermore, we introduce **RECORD**, a novel coordinate-descent-based restoration algorithm that consistently outperforms existing restoration methods by up to 17.8 times. We conduct extensive experiments to assess its compute-performance tradeoff and propose acceleration strategies. |
| title | Rethinking the Vulnerability of Concept Erasure and a New Method |
| topic | Machine Learning Artificial Intelligence Cryptography and Security |
| url | https://arxiv.org/abs/2502.17537 |