(Full PyTorch Code Reference: https://github.com/acids-ircam/diffusion_models)

https://drive.google.com/drive/folders/1nF4LtKfwbiiVLOVonFXAkghj5awSOzvu

Background

• Markov Chain → probability of specific state (t+1) only depends on the current state (t)

$$ p[s_{t+1}|s_t]=p[s_{t+1}|s_1,...,s_t] $$

• Normalizing Flow

• Overview of generative models

• Generative model: Latent variable(잠재 변수) model

→ Goal: Converting a very simple distribution (z) into a specific distribution

→ Most generative models learn the ability to obtain latent variable (z) from given data and transform it

• VAE

→ Mapping to the distribution of specific patterns through the learned decoder network