Speaker:
Pierre Carmier
Title:
Application of information theory to hidden Markov models
Date: November 16th, 2015
Time: 03:30 PM
Place: Showroom
Abstract: This is a subject I have recently become interested in, and the purpose of the talk will therefore be to share my progress on this subject with you and to provide some perspective on what I have understood (and achieved) so far. It will be a blackboard presentation in order to make it pedagogical and also to make some of the calculations easier to follow.
I will begin with a gentle introduction to some of the concepts of information theory which are useful in the context of Bayesian inference. I will continue by presenting a natural strategy one can follow when one seeks to optimize experimental observations of a stochastic system. I will provide along the way several illustrative examples where this strategy can be carried (almost) analytically. Finally, some of the issues related to this approach will be discussed.