15th EUROPT Workshop on Advances in Continuous Optimization

Montréal, Canada, 12 — 14 juillet 2017

15th EUROPT Workshop on Advances in Continuous Optimization

Montréal, Canada, 12 — 14 juillet 2017

Horaire Auteurs Mon horaire
Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402

Advances in Derivative-free and Simulation-based Optimization

13 juil. 2017 09h45 – 11h00

Salle: Nancy et Michel-Gaucher

Présidée par Juliane Mueller

2 présentations

  • Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402
    09h45 - 10h10

    Direct Search Algorithms for solving optimization models with equality constraints

    • Ubaldo Garcia Palomares, prés., Universidade de Vigo

    Well known Direct Search Algorithms (DSA) for solving optimization models have been suggested in the open literature; however, only in the last two years DSA that solve a model with equality constraints have appeared. This talk reviews several approaches for solving models with equality constraints:
    a) Linear equalities are active additional variables at all iterations of the algorithm.
    b) Linear (and nonlinear) equalities constraints are:
    b1) transformed to inequalities with a trivial change to the objective function, or
    b2) a penalization to the objective function.
    Numerical experiments are reported to show the benefits –and shortcomings- of the different approaches.
    This work has been funded by the Galician Government under grants ED341D R2016/012.

  • Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402
    10h10 - 10h35

    Multi-fidelity optimization with computationally expensive black-box functions

    • Juliane Mueller, prés., Lawrence Berkeley National Lab

    We discuss our most recent algorithm developments in derivative-free multi-fidelity optimization where both fidelity models are black-box simulations. The low-fidelity function evaluates faster than the high-fidelity function, but may still require several minutes per evaluation. We use the low-fidelity information to decide where the high-fidelity function should be evaluated. This reduces the number of expensive high-fidelity evaluations we have to do for finding the optimum. We develop sampling strategies that rely on local rather than global correlations for making sampling decisions.

Retour