CARE：面向扩散模型的条件感知表示正则化
原标题：CARE: Condition-Aware Representation Regularization for Diffusion Models
AI 摘要
该论文提出 CARE（Condition-Aware Representation Regularization），一种轻量级即插即用的正则化框架，根据条件相似度动态调整扩散模型的特征分布，无需显式对齐损失或外部监督。在 ImageNet 类到图任务上，CARE 在 40 万训练步内将 FID 降低 19.08%，实现 3.5 倍加速；在文到图任务上，20 万次迭代将 FID 降低 16.61%，并提升语义对齐。CARE 还可与现有正则化方法结合获得额外收益。
正文节选
CARE: Condition-Aware Representation Regularization for Diffusion Models Abstract Recent advances in diffusion models highlight the importance of representation regularization for improving sample quality and training efficiency. However, commonly used regularization methods often overlook the built-in conditions (such as labels or texts) which directly determine the generation target. In this work, we demonstrate how conditioning signals affect the feature distribution and introduce the CARE (C