Plenary Talk: Understanding Image and Physics AI Generation with Harmonic Analysis (video)

Data generation with score diffusions and deep networks can synthesize high quality images, videos, and simulate complex physics. They learn and sample high dimensional probability distributions. Are they memorizing or generalizing? How can they circumvent the curse of dimensionality? What are the underlying structures which are learned? We demonstrate that they do generalize, and that multiscale organizations play a central role. We show that deep networks estimations involve sparse representations, with harmonic analysis transformations adapted to the image geometry.