3 présentations

• 
  09h00 - 09h25

  Stopping.jl : A framework to implement iterative optimization algorithms.

  • Tangi Migot, prés., Univeristy of Guelph
  • Jean-Pierre Dussault, Université de Sherbrooke
  • Samuel Goyette, Université de Sherbrooke

  Due to the increasing need for sophisticated algorithms to solve optimization problems, the reusability of existing codes has become an important question for researchers and practitioners. In this talk, we present the Stopping package and initiate the discussion of an algorithmic framework for (iterative) optimization algorithms. Illustrations of this framework will include an active-set algorithm and a regularization-penalization-active set algorithm for MPCC.
• 
  09h25 - 09h50

  A Julia module for polynomial optimization with complex variables applied to optimal power flow

  • Julie Sliwak, prés., RTE
  • Miguel F. Anjos, GERAD, Polytechnique Montréal
  • Lucas Létocart, LIPN
  • Jean Maeght, RTE
  • Manuel Ruiz, RTE
  • Emiliano Traversi, LIPN

  There is currently no tool for polynomial optimization with complex variables, therefore such problems are usually directly converted to real variables. We propose a Julia module representing polynomial problems in complex variables. This module is applied to optimal power flow problems that naturally involve complex variables due to the alternating current.

  Keywords: Complex Variables, Julia language, Optimal Power Flow in Alternating Current, Polynomial optimization
• 
  09h50 - 10h15

  Tulip.jl: An interior-point solver with abstract linear algebra

  • Mathieu Tanneau, prés.,
  • Miguel F. Anjos, GERAD, Polytechnique Montréal
  • Andrea Lodi, Polytechnique Montreal

  Interior-point methods’ remarkable performance stems from strong algorithmic foundations and efficient linear algebra. Motivated by structured linear programs that arise in decomposition methods, we develop a generic interior-point solver with abstract linear algebra. Computational results demonstrate that, combined with specialized factorization routines, our implementation outperforms state-of-the-art commercial solvers.