Diffusion Seminar

This chapter develops the variational view of diffusion models, beginning with the variational autoencoder (VAE) and building up to the denoising diffusion probabilistic model (DDPM).

This chapter traces the score-based view of generative modeling, moving from energy-based models (EBMs) to score-based generative models.

This chapter presents the Score SDE framework that underpins modern diffusion models, casting the forward and reverse processes as stochastic differential equations with a matching probability-flow ODE.

This chapter introduces the flow-based view, connecting normalizing flows and neural ODEs to the flow matching framework.

This chapter brings the variational, score-based, and flow-based perspectives together, showing how they share a common backbone and are largely equivalent.

This chapter covers how to steer diffusion models toward desired outputs.

This chapter studies advanced numerical solvers that accelerate sampling from diffusion models.

This session covers two complementary routes toward fast diffusion-based generators.