Featured image of post SCORE-BASED GENERATIVE MODELING THROUGH STOCHASTIC DIFFERENTIAL EQUATIONS

SCORE-BASED GENERATIVE MODELING THROUGH STOCHASTIC DIFFERENTIAL EQUATIONS

Score based generative modeling with SDEs

diffusion process可以被表述为以下形式

Untitled

reverse-time SDE可以被表述为以下形式(可以看到需要知道分布分数s_theta)

Untitled

estimating scores for the SDE

和SMLD那篇文章一样,用denoising score matching的方式训练:

Untitled

VE,VP SDEs and Beyond

这里讲了SMLD,DDPM和SDE的关系(SMLD和DDPM可以看作离散的SDEs的两种不同模式)。

对SMLD(Variance Exploding SDE):

Untitled

对DDPM(Variance Preserving SDE):

Untitled

Solving the reverse SDE

作者提出了三种采样方式

general-purpose numerical SDE solvers(reverse diffusion samplers)

DDPM的采样方法

Untitled

被称之为祖先采样(ancestral sampling),而作者提出了reverse diffusion samplers

Untitled

可以证明,ancestral sampling,当beta_i趋近于0的时候,可以转化为reverse diffusion samplers的形式

Predictor-corrector samplers

Untitled

probability flow

对于每个SDE,存在一个确定性的diffusion过程:ODE

Untitled

ODE速度更快但是生成的质量较差。

controallable generation

懒得看了

Built with Hugo
Theme Stack designed by Jimmy