5 Matching Annotations
- Aug 2024
-
lilianweng.github.io lilianweng.github.io
-
the reverse conditional probability is tractable when conditioned on x0:
已知 \(x_0\) 的情况下，反向条件概率可导？为什么
-
Unfortunately, we cannot easily estimate q(xt−1|xt) because it needs to use the entire dataset and therefore we need to learn a model pθ to approximate these conditional probabilities in order to run the reverse diffusion process
因为无法获得全部 dataset，因此需要一个模型来估计真实样本的概率分布，从 \(x_t\) 通过模型预测 \(x_{t-1}\)
-
Note that if βt is small enough, q(xt−1|xt) will also be Gaussian
why?
-
- Mar 2024
-
loop.houmin.site loop.houmin.site
-
在 backward pass 需要的 FLOPs 数大致是 forward pass 的 2 倍
-
- Jan 2023
-
sspai.com sspai.com
-
批注功能支持 Markdown 语法、LaTeX 公式
测试 Markdown 语法： $$x = {-b \pm \sqrt{b^2-4ac} \over 2a}$$.
- ~~strike through~~
-