arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2508.01151cs.CVcs.AI
收藏

面向文本到图像扩散模型的个性化安全对齐

Personalized Safety Alignment for Text-to-Image Diffusion Models

National University of Singapore(新加坡国立大学);Peking University(北京大学);Yale University(耶鲁大学);Collov Labs(Collov实验室)

Yu Lei, Jinbin Bai, Qingyu Shi, Aosong Feng, Hongcheng Gao, Xiao Zhang, Rex Ying

AI总结:

本文提出个性化安全对齐框架PSA,通过用户条件适应提升文本到图像扩散模型的安全性与生成质量,实现安全与质量的动态平衡。

摘要:

文本到图像扩散模型已彻底改变了视觉内容生成,但其部署受到根本性限制的阻碍:安全机制强制执行刚性的、统一的标准,无法反映由年龄、文化或个人信仰塑造的多样化用户偏好。为了解决这一问题,我们提出了个性化安全对齐(PSA),一个将生成安全从静态过滤过渡到用户条件适应的框架。我们引入了Sage,一个大规模数据集,捕捉了1000个模拟用户档案中的多样化安全边界,涵盖了传统数据集常遗漏的复杂风险。通过参数高效的交叉注意适配器整合这些档案,PSA动态调节生成以适应个体敏感性。广泛的实验表明,PSA实现了校准的安全性与质量的权衡:在宽松的档案下,它放松过于谨慎的约束以提高视觉保真度,而在严格档案下,它实施了最先进的抑制,显著优于静态基线。此外,PSA在指令遵循方面优于提示工程方法,确立了个性化作为创建适应性、以用户为中心和负责任的生成AI的关键方向。我们的代码、数据和模型在https://github.com/M-E-AGI-Lab/PSAlign上公开可用。

英文摘要:

Text-to-image diffusion models have revolutionized visual content generation, yet their deployment is hindered by a fundamental limitation: safety mechanisms enforce rigid, uniform standards that fail to reflect diverse user preferences shaped by age, culture, or personal beliefs. To address this, we propose Personalized Safety Alignment (PSA), a framework that transitions generative safety from static filtration to user-conditioned adaptation. We introduce Sage, a large-scale dataset capturing diverse safety boundaries across 1,000 simulated user profiles, covering complex risks often missed by traditional datasets. By integrating these profiles via a parameter-efficient cross-attention adapter, PSA dynamically modulates generation to align with individual sensitivities. Extensive experiments demonstrate that PSA achieves a calibrated safety-quality trade-off: under permissive profiles, it relaxes over-cautious constraints to enhance visual fidelity, while under restrictive profiles, it enforces state-of-the-art suppression, significantly outperforming static baselines. Furthermore, PSA exhibits superior instruction adherence compared to prompt-engineering methods, establishing personalization as a vital direction for creating adaptive, user-centered, and responsible generative AI. Our code, data, and models are publicly available at https://github.com/M-E-AGI-Lab/PSAlign.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑