Seminar 2017.2
August 22, 2017 · 10:00
Quasi-Newton methods for convex constrained nonlinear systems.
Fabrícia Rodrigues de Oliveira · PhD student, IME/UFG
In this talk, we will present a new linesearch Quasi-Newton methods. These methods are based on a proper use of the projection map onto the constraint set and on a derivative-free and nonmonotone linesearch strategy.
August 29, 2017 · 10:00
Pointwise and ergodic convergence rates of a variable metric proximal ADMM
Max Leandro Nobre Gonçalves · IME/UFG
In this talk, we discuss pointwise and ergodic convergence rates for a variable metric proximal alternating direction method of multiplicas (VM-PADMM) for solving linearly constrained convex optimization problems. The VM-PADMM can be seen as a class of ADMM variants, allowing the use of degenerate metrics (defined by noninvertible linear operators). We first propose and study nonasymptotic convergence rates of a variable metric hybrid proximal extragradient (VM-HPE) framework for solving monotone inclusions. Then, the convergence rates for the VM-PADMM are obtained essentially by showing that it falls within the latter framework.
September 5, 2017 · 10:00
The Newton’s method for vector fields on Riemannian Manifolds
Fabiana Rodrigues de Oliveira · PhD student, IME/UFG
In this seminar we will present the Newton's method for finding a singularity of a vector field defined on complete Riemannian manifolds, and discuss some ideas about its convergence analysis.
September 12, 2017 · 10:00
Optimization Methods in the Riemannian Context.
Maurício Silva Louzeiro · PhD student, IME/UFG
In this talk, we will present well-known results related to the convergence of convex programming methods, with the aim of minimizing a convex function defined on a complete Riemannian manifold. In addition, we will comment on the importance of the Toponogov Theorem in the convergence analysis of some methods.
September 26, 2017 · 10:00
A symmetric version of the generalized alternating direction method of multipliers for two-block separable convex programming
Vando Antônio Adona · PhD student, IME/UFG
In this talk, we will analyze the iteration-complexity of a symmetric version of the generalized alternating direction method of multipliers for two-block separable convex programming with linear equality constraints, which inherits the advantages of the classical alternating direction method of multipliers (ADMM), and which extends the feasible set of the relaxation factor of the generalized ADMM. More specifically, we will show that the symmetric ADMM is an instance of a hybrid proximal extragradient framework.
Reference:
[1] Jing Liu, Yongrui Duan and Min Sun. A symmetric version of the generalized alternating direction method of multipliers for two-block separable convex programming. Journal of Inequalities and Applications, (1) 2017:129, 2017.
October 3, 2017 · 10:00
A regularized proximal point method and its iteration complexity
Jefferson D. Gonçalves de Melo · IME/UFG
In this talk, we present a regularized version of the proximal point method which has pointwise iteration complexity O(1/p)log(1/p), improving considerably the O(1/p^2) of the standard proximal point method.
October 19, 2017 · 8:00
On the splitting optimization problem with enlargement
Reinier Díaz Millán · IFG
Given a Hilbert space $\HH$ and a convex functions $f:\dom f\subseteq \HH \rightarrow \R$ and $g:\dom g\subseteq \HH \rightarrow \R$, the Splitting Constrained Optimization Problem ({\emph SCOP}) consist in:
\begin{equation} \min_{x\in C} \, f(x) + g(x), \end{equation} where $C\subseteq \HH$ is a convex and closed set. In this paper, we present two inexact algorithms for solving SCOP. Due to the complexity,in some cases, of computing the proximal operator or the resolvent operator, is imperative to consider relaxed procedures to compute this operator. Our scheme follows the ideas of the $\epsilon-$subdifferential and the inexact proximal point method. The convergence analysis for different stepsizes are presented and some numerical expereminets show the behaviour of our algorithm.
October 26, 2017 · 8:00
On a Gauss-Newton method for solving bound-constrained underdetermined nonlinear systems
Fabrícia Rodrigues de Oliveira · PhD student, IME/UFG
In this talk, we will present a method for solving bound-constrained underdetermined nonlinear systems.We also will discuss results on the global convergence of the method.
November 9, 2017 · 8:00
Proximal alternating direction method of multipliers with larger step size for convex programming with separable structure
Vando Antônio Adona · PhD student, IME/UFG
In this talk, we will present pointwise and ergodic convergence rates for the proximal Alternating Direction Method of Multipliers with larger step size. This version proximal ADMM extends the feasible set, where the update step is chosen, to the interval (0, 2). Some numerical experiments are given to demonstrate the effciency and the performance of the method.
November 16, 2017 · 8:00
Variante Konnov do Algoritmo HPE
Paulo César da Silva Júnior · PhD student, IME/UFG
Inspirados nas ideias de Igor V. Konnov e Benar F. Svaiter - "A combined relaxation method for variational inequalities with nonlinear constraints", publicado em Mathematical Programming 80 (1998), e "A Class of Fejér Convergent Algorithms, Approximate Resolvents and the Hybrid Proximal-Extragradient Method" publicado em J Optim Theory Appl (2014) - propomos um novo algoritmo para o problema de achar zero de operadores monótonos maximais. Mostramos que a sequência gerada é Fejér convergente ao conjunto solução, sempre que este seja não-vazio. Se esta sequência for finita, então, o último ponto calculado é uma solução aproximada do problema, e no caso de a sequência ser infinita, teremos duas situações, a saber, se o tamanho do passo for limitado inferiormente por uma constante positiva, então a sequência converge a uma solução; no outro caso, todo ponto de acumulação da sequência gerada é solução aproximada do problema.
November 23, 2017 · 8:00
Método do ponto proximal para funções localmente Lipschitz em otimização multiobjectivo
Lucas Vidal de Meireles · PhD student, IME/UFG
Nesta exposição, apresentaremos um método do ponto proximal para programação multiobjectivo em que a função vetorial é localmente Lipschitz. Faremos a análise de convergência, provando que todo ponto de acumulação da sequência gerada por este algoritmo, caso exista, é Pareto crítico da função objetivo. Além disso, sob hipótese de convexidade garantiremos convergência a Pareto fraco.
November 30, 2017 · 14:00
Functions defined in Riemannian Manifolds - examples
Maurício Silva Louzeiro · PhD student, IME/UFG
In this talk will be presented special examples of functions defined in Riemannian Manifolds with curvature 0 and with curvature bounded lower by a negative constants. In particular, we will show that by introducing a suitable Riemannian metric, a Euclidean optimization problem having objective function non-convex with gradient non-Lipschitz can be transformed into Riemanian optimization problem with objective convex function with gradient Lipschitz.
December 7, 2017 · 13:00
Iteration-complexity of a Jacobi-type non-Euclidean ADMM for multi-block linearly constrained non-convex programs
Renato D. C. Monteiro · Georgia Institute of Technology, USA
This talk presents the iteration-complexity of a Jacobi-type non-Euclidean proximal al- ternating direction method of multipliers (ADMM) variant for solving (i.e., finding approximate stationary points of) multi-block linearly constrained non-convex programs.
Since the subproblems of this ADMM variant can be solved in parallel, the variant has great potential to solve large scale multi-block linearly constrained non-convex programming instances. Also, our analysis allows the Lagrange multiplier to be updated with a relaxation parameter in the interval (0, 2), and hence extends other previous works which only allow the parameter to chosen in the interval (0,1.618).