4 présentations

• 
  10h30 - 10h55

  Revisiting Optimization Algorithms for Maximum Likelihood Estimation

  • Mai Anh Tien, prés., Université de Montréal
  • Fabian Bastin, Université de Montréal
  • Toulouse Michel, Oklahoma State University

  The development of choice models requires more efficient algorithms for the parameter estimation of choice models. One important estimator is the maximum likelihood estimator. My research examines a range of linear search and trust region methods with standard secant updates (SR1 and BFGS) and statistical approximations (BHHH) for this estimator. New approaches have been developed that obtain better determination of step and Hessian approximation at each iteration. The efficiency of these approaches has been tested with real data of choice models.

• 
  10h55 - 11h20

  Choix adaptatif en micro-simulation

  • Léonard Ryo Morin, prés., Université de Montréal
  • Fabian Bastin, Université de Montréal

  Ma thèse de maîtrise consiste à valider le choix de route adaptatif dans la micro-simulation de réseaux routiers. Le choix de route adaptatif est l'idée de refaire un choix de route à chaque fois qu'un utilisateur atteint un sommet du réseau afin de mieux représenter le comportement des utilisateurs. La norme dans le domaine actuellement est de simplement déterminer un chemin au départ et de le suivre peu importe les conditions rencontrées lors du parcours. Pour valider les résultats de cette hypothèse,le réseau routier de la ville de Namur, en Belgique, sera modélisé.

• 
  11h20 - 11h45

  A Dynamic Formulation for Car Ownership Modeling

  • Renting Xu, University of Maryland
  • Cinzia Cirillo, University of Maryland
  • Fabian Bastin, prés., Université de Montréal

  Most discrete choice models are developed in a static context. We here analyze the impact of
  technological changes on the dynamic of consumer demand, explicitly modeling market evolution and
  accounting for consumers' expectations of future product quality. The decisions timing is formulated as a regenerative optimal stopping problem.

• 
  11h45 - 12h10

  Mixed-Logit Network Pricing

  • François Gilbert, prés., Polytechnique Montréal
  • Patrice Marcotte, Université de Montréal
  • Gilles Savard, Polytechnique Montréal

  In this talk, we address a network pricing problem where users are assigned to the paths of
  a transportation network according to a mixed logit model, i.e., price sensitivity is not assumed
  to be uniform throughout the user population. We propose algorithms based on combinatorial
  approximations and show that the smoothing eect induced by both the discrete choice and
  price sensitivity features of the model actually reduces the number of local optima, and makes
  it easier to obtain a global solution, compared to simpler models where the combinatorics is
  predominant. Also, we estimate the proportion of revenue raised from the various population
  segments, an information that can be used for policy purposes.