Seminários em Análise Convexa e Otimização
Inexact Dual Dynamic Programming
Vincent Guigues (EMAp/FGV-Rio)
Resumo: We introduce an extension of the Dual Dynamic Programming method to solve convex optimization problems.
We call Inexact DDP (IDDP) this extension which applies to situations where all primal and dual subproblems are solved with a bounded error. We also study the convergence of this inexact variant of DDP.
As a by-product, our analysis provides descriptions of the epsilon-subdifferential of the value function of a convex problem when the corresponding primal and dual problems are solved approximately.
The algorithm and convergence analysis can be extended to the stochastic case.
Maiores informações: www.mtm.ufsc.br/~maicon/seminar
Data – Hora: 01/12/2017 – 14h00m
Local: Sala 202 do Departamento de Matemática