Seminario Geometría, Mecánica y Control
Conditional stability of the geodesic Euler method on Riemannian manifolds with applications to neural network architectures
Ponente: Brynjulf Owren (NTNU, Norway)Fecha: jueves 04 de junio de 2026 - 12:00Lugar: Aula Gris 3, ICMAT
Resumen:
Numerical integrators for differential equations on manifolds are well established, but their stability and convergence properties are less developed. In a recent paper, we found accurate bounds on the time step for the geodesic Euler method to be non-expansive in the case of constant curvature spaces. We will briefly review these results and also discuss some work in progress on neural network architectures that admit non-expansive layer maps on Hadamard manifolds.
