This post discusses a paper that argues diffusion models, which typically work by denoising noisy data to generate new images, can be understood as a type of evolutionary algorithm. The authors break down how both diffusion models and evolutionary algorithms involve iterative data updates and sampling from complex
Sort: