Complexidade por iteração do método HPE e sua versão acelerada para otimização convexa

• Autor(a) principal: Chagas, Marcus Vinícius de Morais
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFG
• dARK ID: ark:/38995/0013000003p64
• Texto Completo: http://repositorio.bc.ufg.br/tede/handle/tede/12854
• topic: Otimização convexa; Zeros de operadores monótonos maximais; Método do ponto proximal; Método HPE; Método A-HPE; Método acelerado de primeira ordem; Complexidade por iteração; Convex optimization; Zeroes of maximal monotinic operators; Proximal method point; HPE methods; Accelerated first order method; Iteration-complexity; CIENCIAS EXATAS E DA TERRA::MATEMATICA

In this work, we analyze the Hybrid Proximal Extragradiente (HPE) method to find zeroes of maximal monotone operators and its accelerated version Accelerated Hybrid Proximal Extragradient (A-HPE) to solve convex optimization problems whose objective function is given by the sum of two other convex functions, one differentiable with Lipschitz gradient and another one not necessarily differentiable. The HPE method was introduced by Solodov and Svaiter, it consists of an inexact version of the proximal point method having its proximal subproblems inexactly solved using a relative error criterion followed by an extragradient step. The HPE can also be seen as a framework, in the sense that many other methods for minimizing convex functions and more generally to find zeroes of maximal monotone operators can be seen as instances of the HPE method, such as the extragradient method, regularized Newton type method, ADMM, etc. In this work, we will analyze both the asymptotic convergence of the HPE method and its iteration-complexity.We will also analyze the iteration-complexity of the A-HPE method proposed by Monteiro and Svaiter. The A-HPE is a first-order accelerated method, i.e., it uses only information of the functional values and the first derivative or subgradients of the objective function and has optimal iteration-complexity.