Seminario de Matematica Aplicada: New perspectives for increasing efficiency of minimization schemes

06/11/2015 16:48

Resumo: In this talk we prove a new complexity bound for a variant of Accelerate Coordinate Descent Method. We show that this method always outperform the standard Fast Gradient Methods on many optimization problems with dense data. As application examples, we consider unconstrained convex quadratic minimization problem, and the problems arising in Smoothing Technique. On some special problem instances, the provable acceleration factor can reach the square root of number of variables

Palestrante: Yurii Nesterov, Center for Operations Research and Econometrics, Universidade Católica de Louvain, Bélgica
Data: 09/11/15
Horário: 14:00 às 14:50h
Local: Auditório (LAED) do Departamento de Matemática, andar térreo