Full Professor, Ecole Normale Superieure
1 paper at NeurIPS 2025
Implicit dynamical regularization during training gives diffusion models a generalization window that widens with the training set size, so stopping within this window prevents memorization.
