4 présentations

• 
  15h30 - 15h55

  Estimation et test d'adequation pour des modèles de copules à changement de régime, avec application

  • Mamadou Yamar Thioub, prés., HEC Montreal

  Dans cette présentation, j’exposerai différents aspects de la modélisation de la dépendance de séries temporelles univariées, par des modèles de copule avec changement de régime. Afin de faciliter l’utilisation des méthodologies, je présenterai les fonctionnalités de la librairie HMMcopula, qui a été développée pour ces modèles, et qui disponible sur CRAN.

  Test d'adéquation, séries temporelles, copules, modèles dynamiques
• 
  15h55 - 16h20

  Multiscale Gaussian process regression for feature enhancement in laser-based bathymetric SLAM

  • Thomas Hitchcox, prés., McGill University
  • James Richard Forbes, McGill University

  This presentation will describe the application of Gaussian process regression to bathymetric SLAM, whereby two rounds of hyperparameter optimization are performed to extract and upsample keypoint sets from sparse laser scans of a subsea environment. Results from real data will be presented, including 3D scans of the Sweepstakes heritage shipwreck.

  Keywords: bathymetric SLAM, hyperparameter optimization, point clouds
• 
  16h20 - 16h45

  Synthesis of optimal yet robust controllers for negative imaginary systems

  • Ken Lee, prés., McGill University

  There is an input-output stability theorem for a class of systems called negative imaginary. Such a system can be stabilized by strictly negative imaginary controllers. By exploiting the properties of negative imaginary systems, the optimal yet robust controller synthesis problem can be formulated as an optimization problem using linear matrix inequality constraints.

  robust control, optimal control, negative imaginary systems
• 
  16h45 - 17h10

  System identification and optimal control of a fatigue testing rig for aircraft

  • James Richard Forbes, prés., McGill University
  • Robyn Fortune, McGill University - Graduate Student

  In this work, a technique called system identification was implemented in order to "identify" numerical models using data. The system ID methods involve a least-squares approach. They were applied to fatigue test rig data from the National Research Council. The resulting model was used to synthesize H-infinity-optimal controllers using linear matrix inequalities and convex optimization.

  Keywords: least squares, convex optimization, control