ENSEEIHT/projet-fin-etude-rapport
1
0
Fork 0
Code Issues Pull requests Packages Projects Releases Wiki Activity

ddpm cfg algo requirements fix

Browse source
This commit is contained in:
Laureηt 2023-08-28 16:46:34 +02:00
parent 2a0064b09a
commit bec9e7ab0f
Signed by: Laurent
SSH key fingerprint: SHA256:kZEpW8cMJ54PDeCvOhzreNr4FSh6R13CMGH/POoO8DI
1 changed files with 2 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
pdf/paper.tex
View file

@ -710,10 +710,10 @@ Formellement, l'algorithme d'entraînement par \gls{cfg} est décrit dans l'algo
\begin{algorithmic}[1] \begin{algorithmic}[1]
\Require $T \in \mathbb{N}^\star$: number of diffusion steps \Require $T \in \mathbb{N}^\star$: number of diffusion steps
\Require $\alpha_t \in \mathbb{R}^T$: variance schedule \Require $\alpha_t \in \mathbb{R}^T$: variance schedule
\Require $p_\text{uncond} \in [0, 1]$: probability of unconditional training
\Require $\boldsymbol{x}_0$: data distribution to be learned \Require $\boldsymbol{x}_0$: data distribution to be learned
\Require $\boldsymbol{c}$: embeddings distribution to be learned \Require $\boldsymbol{c}$: embedding distribution to be learned
\Require $\epsilon_\theta$: neural network \Require $\epsilon_\theta$: neural network
\Require $p_\text{uncond}$: probability of unconditional training
\Repeat \Repeat
\State $(x_0, c) \sim (\boldsymbol{x_0}, \boldsymbol{c})$ \State $(x_0, c) \sim (\boldsymbol{x_0}, \boldsymbol{c})$
\State $c \leftarrow \varnothing$ with probability $p_\text{uncond}$ \State $c \leftarrow \varnothing$ with probability $p_\text{uncond}$