Pristine Models：预训练阶段分离安全与不安全概念

AI 摘要

该论文提出 Pristine Models，一种文本到图像生成架构，通过在预训练阶段将安全与不安全概念知识分离到独立参数化的注意力分支中，从而在推理时可直接移除不安全分支。在 ImageNet 和 CC12M 上的实验表明，该方法在减少不安全内容生成的同时，比仅用安全数据训练或事后概念移除方法保留了更高的图像质量和文本对齐度。

正文节选

Pristine Models Abstract Training on broad, unfiltered data enables high-quality image generation, but can also teach models concepts that should not be generated; restricting the training data ensures safety, but reduces quality because examples containing unsafe concepts also provide useful knowledge that is not itself unsafe. Existing approaches largely address this trade-off post hoc through concept unlearning, which can lead to catastrophic forgetting as the number of concepts to remove gro

发布时间：2026-10-14 12:00

抓取时间：2026-10-14 14:39

来源机构：arXiv