What is Sampling?

Under Construction

As said in #what is generating, sampling means to pick data out according to a data distribution; A data distribution is something that assigns probabilities to data, for example a cat image distribution \(p_\text{CatImages}\) would assign high probability to cat images, and low probability to everything else like dog images.

As said in #how 2 diffuse, for diffusion models, sampling is done through solving the corresponding ODE/SDE that would transform an input from an initial distribution, like noise from a gaussian, into a clean image, like a cat image from a cat distribution.

What is a Sampler?

While not all of them were conceived this way, all samplers can be thought of as something that can solve the diffusion ODE/SDE.