Minimization (SAM), a recently proposed gradient-based optimization method [Foret et al., 2021] that significantly improves the generalization of deep neural

While these algorithms have been extensively studied, both theoretically and practically, see, e.g., [