{"id":"arxiv/2609.14141","url":"https://arxiv.org/abs/2609.14141","tags":["math.OC","cs.LG","stat.ML"],"score":0,"title":"Riemannian ascent--descent for nonconvex nonconcave minimax landscapes: convergence to basin saddle points and applications to distributionally robust optimization","domain":"math.OC","source":"arxiv","authors":["Rishabh Dixit","Pranav Upadrashta","Alex Cloninger"],"reviews":0,"subject":"math.OC","summary":"We study a class of distributionally robust optimization (DRO) problems for the statistical risk problem, formulated as minimax problems over the product of a Euclidean space and a Riemannian manifold. Because the resulting minimax landscape is nonconvex nonconcave in general, no globally convergent first order method is known to be available. We instead introduce the notion of a \\emph{basin saddle point}, a Nash equilibrium defined locally on the Cartesian product of a $δ$ basin around a connected component of the local minima critical set and a geodesic ball on the measure manifold. We…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-09-12","reactions":{}}