Departmento de Estadistica e Investigacion Operativa
Universidad Rey Juan Carlos
We devise a methodology that exploits the properties of Krylov methods and negative curvature information of nonlinear nonconvex optimization problems. In doing so, we speed the convergence rate to local minima, avoiding explictly the possibility of being trapped in saddle-points. We give theoretical details of the method and show its performance when applied to various real world examples.