The Effect of Parametrization on Nonconvex Optimization Landscapes - Seminar

Explore a 54-minute seminar by Joe Kileel on "The Effect of Parametrization on Nonconvex Optimization Landscapes" presented at the Instituto de Matemática Pura e Aplicada. Delve into the relationship between paired optimization problems through smooth parametrization of domains, examining how this affects optimization landscapes. Discover a new geometric framework for studying parametrizations and their impact on landscapes, with applications in low-rank matrix and tensor optimization, semidefinite programming, neural network training, and symmetry quotienting. Gain insights from this joint work with Eitan Levin and Nicolas Boumal, presented as part of the Seminário Centro PI series.