Lecturers

Thomas Schön, Uppsala University
Web page

Schedule and place

This 15-hour course will take place in 6 sessions over four days on October 3, 4, 5, 6, 2017.

Schedule :

  • October 3d and 5th : from 9:30 to 12:30 and from 13:30 to 16:30 (including a 30-minute coffee break in each session).
  • October 4th and 6th : from 9:30 to 12:30.

The course will take place at the Green Room, ground floor of the U-residence, Generaal Jacqueslaan 271, 1050 Brussels.
See http://www.u-residence.be/en/location

Description

The aim of this course is to provide an introduction to the theory and application of computational methods for inference in nonlinear dynamical systems, as well as more general problems. More specifically we will introduce sequential Monte Carlo (SMC) methods and Markov chain Monte Carlo (MCMC) methods and show how these can be used to solve challenging learning (system identification) and state estimation problems in nonlinear dynamical systems. We will also briefly discuss SMC in a more general context, showing how it can be used as a generic tool for sampling from complex probability distributions.

List of topics:

  • Probabilistic modeling of dynamical systems
  • Filtering and smoothing
  • The Monte Carlo idea and importance sampling
  • Particle filtering / Sequential Monte Carlo
  • Basic convergence theory for particle filters
  • Likelihood estimation and maximum likelihood parameter learning
  • Bayesian parameter learning
  • Particle Markov Chain Monte Carlo
  • Generic SMC / SMC Samplers

Prerequisites: Basic understanding of probability theory, statistics and numerical computations.

Course material

The course is based on the manuscript: Thomas B. Schön and Fredrik Lindsten. Learning of dynamical systems - Particle filters and Markov chain methods, Lecture notes, 2017 and various recent papers. Link to the lecture notes: http://www.it.uu.se/research/systems_and_control/education/2017/smc/SMC2017.pdf

Slides for the lectures will be made available online.

Evaluation

Either via paper or a Google form.