ProfOptimization2016

Seminar 2021.2

Organizer: Orizon P. Ferreira. Talks were held online via Google Meet.

December 9, 2021

The DC Algorithm on Hadamard manifolds

Elianderson Meneses Santos · PhD student, IME/UFG

The DC Algorithm (DCA), proposed originally by Tao and Souad in [Algorithms for solving a class of nonconvex optimization problems. Methods of subgradients. North-Holland Mathematics Studies, vol.129, pp. 249-271.North-Holland, Amsterdam (1986)], was the first method devoted to solving DC problems, that is, problems that require minimizing an objective function defined as a difference of two convex functions. Over last 35 years, several versions of the DCA for solving practical problems that can be modeled as DC have been developed and, in addition, several other methods for DC programming have emerged from the DCA ideas, e.g. the Proximal DCA (PDCA), and the Boosted DCA (BDCA). In this seminar we present an extension of the DCA to the context of DC problems on Hadamard manifolds. We show that the method is well defined and, additionally, that every cluster point of the sequence generated by the DCA is a critical point of the considered problem, which retrieves a property of the original DCA in the Euclidean space.

December 23, 2021

Sobre o método BFGS para otimização irrestrita multiobjetivo

Danilo Rodrigues de Souza · PhD student, IME/UFG

Nesta apresentação, propomos uma extensão do método BFGS para otimização multiobjetivo que, assim como o clássico, usa busca linear de Wolfe para obter os comprimentos de passo. Semelhante ao caso escalar, se os comprimentos de passos satisfazem as condições de Wolfe multiobjetivo, então a atualização proposta se mantém definida positiva. A boa definição do método é obtida mesmo quando as funções objetivo são não-convexas. Usando hipóteses de convexidade, estabelecemos convergência superlinear para pontos Pareto ótimos. As hipóteses consideradas são extensões diretas das usadas no método BFGS escalar. Discutiremos também estratégias de globalização do método BFGS multiobjetivo para o contexto não-convexo.

January 20, 2022

Computing inexact K-steepest descent directions and a new line search for Vector Optimization

Flávio Pinto Vieira · PhD student, IME/UFG

In this talk, two results are presented. In 2004, Iusem and Graña Drummond introduced the concept of $\sigma-$approximate $\mathcal{K}-$steepest descent direction. They showed that by replacing the Cauchy direction with these directions, the convergence result of the generated sequence is the same: every accumulation point is critical. We will present an efficient procedure for computing these directions when the cone $\mathcal{K}$ is finitely generated. Yunda Dong, in 2010 and 2012, introduced a new linear search procedure for Conjugated Gradient methods using only first-order information, i.e., without working with functional values. We extended his works to Vector Optimization. We studied conjugate gradient methods, showing convergence, when the following $\beta_k'$s are used: Fletcher-Reeves, conjugate descent, Dai-Yuan, Polak-Ribière-Polyak, and Hestenes-Stiefel.

January 27, 2022

A Semilocal Convergence of a Secant–Type Method for Solving Generalized Equations

Paulo César da Silva Júnior · PhD student, IME/UFG

In this seminar will be presented the paper "A Semilocal Convergence of a Secant–Type Method for Solving Generalized Equations (https://link.springer.com/article/10.1007/s11117-006-0044-3)".

"Abstract: In this paper we present a study of the existence and the convergence of a secant–type method for solving abstract generalized equations in Banach spaces. With different assumptions for divided differences, we obtain a procedure that have superlinear convergence. This study follows the recent results of semilocal convergence related to the resolution of nonlinear equations (see https://www.sciencedirect.com/science/article/pii/S0898122102001475)"

February 3, 2022

Newton-type Methods for Multiobjective Optimization and a Study of Conjugate Gradient Methods for Vector Optimization

Fernando Santana Lima · PhD student, IME/UFG

In this we propose two Newton-type methods for solving (possibly) nonconvex unconstrained multiobjective optimization problems. The first is directly inspired by the Newton method designed to solve convex problems, whereas the second uses second-order information of the objective functions with ingredients of the steepest descent method. Also we propose a general conjugate gradient (CG) method for vector optimization problems with sufficient descent property regardless of linear search and of parameter choices

February 10, 2022

Some Recent Advances in Optimization on Riemannian Manifolds

Glaydston de Carvalho Bento · IME/UFG

In this talk, we will discuss the existence of solutions to equilibrium problems associated with pseudomonotone bifunctions, as well as a new proposal of resolvent from which it is possible to guarantee well-definedness of a proximal-type method for equilibrium problems on Hadamard manifolds. Our approach requires the study of some elements of convex analysis which will also be discussed.

February 17, 2022

An Inexact Proximal Point Method for Variational Inequality on Hadamard Manifolds

Erik Alex Papa Quiroz · IME/UFG

In this talk we present an inexact proximal point method for variational inequality problem on Hadamard manifolds and study its convergence properties. The proposed algorithm is inexact in two senses. First, each proximal subproblem is approximated by using the enlargement of the vector field in consideration and then the next iterated is obtained by solving this subproblem allowing a suitable error tolerance. As an application, we obtain an inexact proximal point method for constrained optimization problems, equilibrium problems and nonlinear optimization problems on Hadamard manifolds.

February 24, 2022

Empacotamento de chapéus esféricos em superfícies de toros planares: uma aplicação em quantização vetorial

Fabiano Boaventura de Miranda · UEG

O processo de quantização é utilizado em muitas aplicações que envolvem a conversão de informação de formato analógico para digital, sobretudo em cenários envolvendo variáveis contínuas. A técnica consiste em mapear um conjunto de dados de um domínio contínuo (ou muito grande) para um domínio enumerável. O desafio envolve escolher um conjunto de pontos para representar os dados de modo a minimizar a perda geral ocorrida pela compressão da informação. Em aplicações envolvendo a quantização de fontes gaussianas, o desafio pode ser formulado como um problema de otimização que consiste em empacotar chapéus esféricos sobre a superfície de uma esfera Euclidiana de modo a maximizar a distância mínima entre os centros de cada chapéu. O conjunto de soluções desse problema de otimização é denominado código esférico. Apresentaremos um método construtivo, que utiliza um reticulado k-dimensional como pré-imagem do código esférico. O mergulho desse reticulado em dimensão 2k forma um código esférico cujos pontos moram na superfície de um toro planar. Uma das vantagens desse método é reduzir custo computacional envolvido na quantização, utilizando a estrutura de reticulado na metade da dimensão do código e a manipulação de esferas em dimensões mais altas, o que melhora a eficiência em problemas de quantização vetorial, como por exemplo, hiperesferas em dimensão 4 à 48. Diversas construções são apresentadas e os resultados são comparados com os principais quantizadores vetoriais conhecidos na literatura.

March 10, 2022

Convexity of sets and quadratic functions on the hyperbolic space

Orizon Pereira Ferreira · IME/UFG

In this talk some concepts of convex analysis on hyperbolic space are studied.

We first study properties of the intrinsic distance, for instance, we present the spectral decomposition of its Hessian. Next, we study the concept of convex sets and the intrinsic projection onto these sets. We also study the concept of convex functions and present first and second order characterizations of these functions, as well as some optimization concepts related to them. An extensive study of the hyperbolically convex quadratic functions is also presented.

March 17, 2022

Desempenho numérico do método BFGS para Otimização Multiobjetivo

Danilo Rodrigues de Souza · PhD student, IME/UFG

Nesse seminário faremos uma breve apresentação do método BFGS para otimização multiobjetivo, proposto por L. F. Prudente e D. R. Souza, e os testes numéricos realizados. O método proposto tem como principal novidade o uso de passo de Wolfe com a atualização da Hessiana aproximada em todas as iterações. Para fins de comparação, utilizamos um método BFGS multiobjetivo onde o método dá passo satisfazendo Armijo e, se uma cautelosa condição é satisfeita, atualiza a Hessiana aproximada. Para esse mesmo método, também propomos o uso de passo de Wolfe, obtendo assim um terceiro método. Os testes numéricos foram realizados com esses três métodos e os resultados obtidos mostram que o uso do passo de Wolfe com a atualização da Hessiana aproximada em todas as iterações, cria uma potencial vantagem em relação aos outros dois.

March 24, 2022

Computing inexact K-steepest descent directions and a new line search procedure for Vector Optimization

Flávio Pinto Vieira · PhD student, IME/UFG

In this work, we proposes a new linear search and a way for the computation of sigma-approximate direction. Yunda Dong, in 2010 and 2012, introduced a new linear search procedure for Conjugated Gradient methods using only first-order information, i.e., without working with functional values. We extended his works to Vector Optimization. We studied conjugate gradient methods, showing convergence when the following beta_k's are used: Fletcher-Reeves, conjugate descent, Dai-Yuan, Polak-Ribiére-Polyak, and Hestenes-Stiefel. We also use this line search in the gradient method, showing its convergence. In 2004, Iusem and Graña Drummond introduced the concept of sigma-approximate K-steepest descent direction. They showed that by replacing the Cauchy direction with these directions, the convergence result of the generated sequence is the same: every accumulation point is critical. We will present an efficient procedure for computing these directions when the cone K is finitely generated.

March 31, 2022

On the computation of approximated steepest descent direction in vector optimization

Luis Román Lucambio Pérez · IME/UFG

When classical optimization procedures are extended to the context of vector optimization, one of the main issues is how to compute, efficiently, the direction along which the line search will be performed. This talk will be on this theme. Approximated steepest descent directions can be computed via an inclusion convex problem, which can be solved by minimizing a quadratic convex function over a simplex. The proposed procedure has finite termination.