solver-iterates

GLENS is a data-efficient global search method that uses diffusion models to generate diverse, high-quality initial guesses for local minima in non-convex optimization problems by leveraging intermediate solver iterates as free data augmentation.