Information
Title: Understanding diffusion models: A unified perspective
Year: 2022
Authors: Calvin Luo
Publication:
Extra：
Zotero: Luo - 2022 - Understanding diffusion models A unified perspective.pdf
PDF: Luo - 2022 - Understanding diffusion models A unified perspective.pdf
Citation: [1]
C. Luo, “Understanding diffusion models: A unified perspective,” arXiv preprint arXiv:2208.11970, 2022.
Notes
problem
idea
insight
keyword:
Annotations
Luo - 2022 - Understanding diffusion models A unified perspective.pdf
Three Equivalent Interpretations » Page 15
因此，通过预测原始图像 x0 来学习 VDM 等同于学习预测噪声；但根据经验，一些研究发现预测噪声能带来更好的性能。> therefore shown that learning a VDM by predicting the original image x0 is equivalent to learning to predict the noise; empirically, however, some works have found that predicting the noise resulted in better performance [5, 7]. » Page 16
关于DM与score networks等价性证明>
关于DM与score networks等价性证明（续）>
总结三个等价表示> We have therefore derived three equivalent objectives to optimize a VDM: learning a neural network to predict the original image x0, the source noise 0, or the score of the image at an arbitrary noise level ∇ log p(xt). » Page 17