1 paper
Matthew Niedoba, Berend Zwartsenberg, Kevin Murphy +1
We propose a simple, training-free mechanism which explains the generalization behaviour of diffusion models. By comparing pre-trained diffusion models to their theoretically optim…