Seminar 2023.2
October 26, 2023
A projected subgradient method for the computation of adapted metrics for dynamical systems
Erik Alex Papa Quiroz
In this talk, we extend a recently established subgradient method for the computation of Riemannian metrics that optimizes certain singular value functions associated with dynamical systems. This extension is threefold.
First, we introduce a projected subgradient method which results in Riemannian metrics whose parameters are confined to a compact convex set and we can thus prove that a minimizer exists; second, we allow inexact subgradients and study the effect of the errors on the computed metrics; and third, we analyze the subgradient algorithm for three different choices of step sizes: constant, exogenous and Polyak.
November 9, 2023
On the convergence rate of the conditional gradient method
Maurício Silva Louzeiro
In this talk, we discuss convergence rate results related to the scalar conditional gradient method as well as its multiobjective version.
November 16, 2023
On projection mappings and the gradient projection method on hyperbolic space forms
Max Leandro Nobre Gonçalves
In this talk we examine the gradient projection method as a solution approach for constrained optimization problems in $\kappa$-hyperbolic space forms, particularly for potentially non-convex objective functions. We consider both constant and backtracking step sizes in our analysis. Our studies are based on the hyperboloid model, commonly referred to as the Lorentz model. In our investigation, we present several innovative properties of the intrinsic $\kappa$-projection onto convex sets of $\kappa$-hyperbolic space forms. These properties are crucial for analyzing the method and also hold independent significance. We discuss the relationship between the intrinsic $\kappa$-projection and the Euclidean orthogonal projection as well as the Lorentz projection. Moreover, we provide formulas for the intrinsic $\kappa$-projection into specific convex sets using the Euclidean orthogonal projection and the Lorentz projection. Regarding the convergence results of the gradient projection method, we establish two main findings. Firstly, we demonstrate that every accumulation point of the sequence generated by the method with backtracking step sizes is a stationary point for the given problem. Secondly, assuming the Lipschitz continuity of the gradient of the objective function, we show that each accumulation point of the sequence generated by the gradient projection method with a constant step size is also a stationary point. Additionally, we provide an iteration complexity bound that characterizes the number of iterations needed to achieve a suitable measure of stationarity for both step sizes. Finally, we explore the properties of the constrained Fermat-Weber problem, demonstrating that the sequence generated by the gradient projection method converges to its unique solution.
November 30, 2023
The steepest descent method for multicriteria optimization: Towards an add and drop products theory of the multiproduct firm
Tiago Sousa Mota
December 7, 2023
Um algoritmo do ponto proximal para problemas de minimização não-convexos
Claudemir Rodrigues Santiago
Nesta palestra apresentamos o algoritmo bem como alguns resultados do trabalho de Fukushima M. e Mine H.: “A Generalized Proximal Point Algorithm for Certain non-Convex Minimization Problems“. Nessa proposta a função objetivo é a soma de uma função continuamente diferenciável e uma função convexa. A classe de tais problemas contém como caso especial o de minimização de uma função continuamente diferenciável sobre um conjunto convexo fechado. Este método pode ser considerado como uma generalização do algoritmo do ponto proximal para lidar com a não-convexidade da função objetivo linearizando o termo diferenciável em cada iteração.
December 14, 2023
A successive centralized circumcentered-reflection method for the convex feasibility problem
Di Liu · IMPA
In this talk, we present a successive centralization process for the circumcentered-reflection method with several control sequences for solving the convex feasibility problem in Euclidean space. Assuming that a standard error bound holds, we prove the linear convergence of the method with the most violated constraint control sequence. Moreover, under additional smoothness assumptions on the target sets, we establish the superlinear convergence. Numerical experiments confirm the efficiency of our method.
January 11, 2024
Local Convergence of Quasi-Newton Methods under Metric Regularity.
Paulo César da Silva Júnior
The talk considers quasi-Newton methods for generalized equations in Euclidean spaces under metric regularity and provides a sufficient condition for q-linear convergence. Additionally, we demonstrate that the Broyden update satisfies this sufficient condition.
January 18, 2024
Taxa de Convergência de um Algoritmo do Ponto Proximal para Problemas de Minimização Não-Convexos
Claudemir Rodrigues Santiago
Nesta palestra analisaremos os resultados de convergência do trabalho de Fukushima M. e Mine H. “A Generalized Proximal Point Algorithm for Certain non-Convex Minimization Problems“; tal método foi proposto para minimizar uma função objetivo não-convexa que pode ser descrita como a soma de uma função continuamente diferenciável e uma função convexa não necessariamente diferenciável. Apresentaremos o algoritmo e algumas propriedades necessárias para a análise da taxa de convergência. Provaremos que a sequência gerada pelo algoritmo converge linearmente para uma solução do problema sob certas condições. Na convergência do algoritmo, assume-se que a sequência dos parâmetros proximais {c_k} seja somente limitada, porém, na análise da taxa de convergência consideraremos o caso especial em que {c_k} é constante. Por fim, discutiremos algumas possibilidades de alterações nas funções que compõem a função objetivo, além de perspectivas para pesquisas futuras.