Sander Dieleman

A flurry of recent activity in the space of continuous diffusion models for language, after a few years of relative dormancy, suggests that this approach is making something of a comeback.

Sampling from a diffusion model is an iterative process: at each step, the denoiser estimates the tangent direction to a path through input space.

Most contemporary generative models of images, sound and video do not operate directly on pixels or waveforms.

A bit of signal processing swiftly reveals that diffusion models and autoregressive models aren’t all that different: diffusion models of images perform approximate autoregression in the frequency domain!

The noise schedule is a key design parameter for diffusion models.

Diffusion models split up the difficult task of generating data from a high-dimensional distribution into many denoising tasks, each of which is much easier.

Guidance is a powerful method that can be used to enhance diffusion model sampling.

Diffusion models appear to come in many shapes and forms.

Diffusion models have completely taken over generative modelling of perceptual signals such as images, audio and video.

Classifier-free diffusion guidance1 dramatically improves samples produced by conditional diffusion models at almost no cost.