4 présentations

• 
  10h30 - 10h55

  LocalSolver: Black-Box Local Search for Combinatorial Optimization

  • Frederic Gardi, prés., Bouygues e-lab
  • Thierry Benoist, Bouygues e-lab
  • Julien Darlay, Bouygues e-lab
  • Bertrand Estellon, Aix-Marseille Université
  • Romain Megel, Bouygues e-lab
  • Karim Nouioua, Aix-Marseille Université

  We present LocalSolver 2.0, model-and-run solver based on local-search techniques. It can handle very large nonlinear problems with millions of 0-1 decisions. LocalSolver offers simple APIs as well as an efficient modeling language for fast prototyping. It is used in several real-life applications and has succeeded the first tour of Google ROADEF/EURO Challenge.

• 
  10h55 - 11h20

  Recent Developments in Surrogate-Based Algorithms for Constrained Black-Box Optimization

  • Rommel Regis, prés., Saint Joseph's University

  This talk will present a survey of some recent surrogate-based algorithms for the optimization
  of expensive black-box functions subject to expensive black-box constraints. It will also provide numerical results on some test problems and on an automotive problem with 124 decision variables and 68 black-box constraints.

• 
  11h20 - 11h45

  The Mesh Adaptive Direct Search Algorithm with Reduced Number of Directions

  • Charles Audet, GERAD - Polytechnique Montréal
  • Andrea Ianni, Università di Roma La Sapienza
  • Sébastien Le Digabel, prés., GERAD, Polytechnique Montréal
  • Christophe Tribes, Polytechnique Montréal

  The Mesh Adaptive Direct Search (MADS) algorithm is designed for blackbox optimization where the objective function and constraints correspond to a costly computer simulation. Each iteration of the algorithm consists of launching the simulation at a nite number of trial points. These candidates are constructed using typically 2n directions, where n is the number of variables. This presentation shows some ways of reducing that number to a minimal positive spanning set of n + 1 directions. This transformation is generic and can be applied to any method that generates at least n + 1 directions.

• 
  11h45 - 12h10

  Pyomo: Modeling and Solving Mathematical Programs in Python

  • David L. Woodruff, prés., UC Davis
  • Jean-Paul Watson, Sandia National Laboratory
  • William Hart, Sandia National Laboratory

  We describe Pyomo, an open source software package for modeling and solving mathematical programs in Python. Pyomo can be used to define abstract and concrete problems, create problem instances, and solve these instances with standard open-source and commercial solvers. Pyomo provides a capability that is commonly associated with algebraic modeling languages such as AMPL, AIMMS, and GAMS. In contrast, Pyomo’s modeling objects are embedded within a full-featured highlevel programming language with a rich set of supporting libraries. Pyomo leverages the capabilities of the Coopr software library, which together with Pyomo is part of IBM’s COIN-OR open-source initiative for operations research software. Coopr integrates Python packages for defining optimizers, modeling optimization applications, and managing computational experiments. Numerous examples illustrating advanced scripting applications are provided.