Journées de l'optimisation 2017

HEC Montréal, 8-10 mai 2017

1er Atelier Canadien sur l'optimisation des soins de santé (CHOW)

HEC Montréal, 10-11 mai 2017


HEC Montréal, 8 — 11 mai 2017

Horaire Auteurs Mon horaire
Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402

TD4 Apprentissage et diagrammes de décision / Learning and decision diagrams

9 mai 2017 15h30 – 17h10

Salle: Meloche Monnex

Présidée par Andre Augusto Cire

3 présentations

  • Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402
    15h30 - 15h55

    Optimization methods for neural networks training

    • Dimitri Papadimitriou, Présentateur, Bell Labs

    Given a set of labeled data points, the optimization problem associated to the training of neural networks aims at determining the parameters, e.g., synaptic weights, which minimize the empirical loss between the true output to the given input and the predicted output. The (regularized) problem is nonconvex even when the loss (and the regularization) function is convex. We analyze and compare extended bundle and trust region methods for nonconvex loss and non/convex non/smooth regularization term.

  • Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402
    15h55 - 16h20

    A hybrid decision diagram approach for the job shop scheduling problem

    • Jaime Gonzalez, Présentateur, Polytecnique Montréal
    • Louis-Martin Rousseau, Polytechnique Montréal
    • Andre A. Cire, University of Toronto Scarborough
    • Andrea Lodi, Polytechnique Montréal

    We propose an optimization framework which integrates mixed-integer programming (MIP) and multivalued decision diagrams (MDDs) for optimization. A MDD representation of the problem identifies parts of the search space that can be efficiently explored by MIP technology, while the MIP results are iteratively used to refine the MDD.

  • Cal add eabad1550a3cf3ed9646c36511a21a854fcb401e3247c61aefa77286b00fe402
    16h20 - 16h45

    Decompositions based on Decision Diagrams

    • Andre Augusto Cire, Présentateur, University of Toronto
    • David Bergman, University of Connecticut

    This talk describes a new decomposition approach where small-sized decision diagrams exactly represent different portions of a discrete optimization problem, all of which are linked through special constraints. We discuss potential techniques to solve the underlying decomposition problem and show a number of applications of this method.