Skip to main content

Moritz Schauer: Automatic Backward Filtering Forward Guiding for Markov processes and graphical models

Time: Mon 2020-11-09 15.15 - 16.15

Lecturer: Moritz Schauer, Chalmers University of Technology, University of Gothenburg

Location: Zoom, meeting ID: 621 4469 8204

Abstract

 I introduce the automatic Backward Filtering Forward Guiding (BFFG) paradigm for programmable inference on latent states and model parameters in probabilistic graphical models. Specifically, I consider probabilistic graphical models with discrete and continuous time Markov processes as building blocks. Starting point is a generative model, a forward description of the probabilistic process dynamics. In Backward Filtering Forward Guiding the information provided by observations is backpropagated through the model to transform the generative (forward) model into a pre-conditional model guided by the data, a model which approximates the actual (intractable)conditional model with known likelihood-ratio between the two.

This guided generative model can be incorporated in different approaches to efficiently sample latent states and parameters conditional on observations, and is suitable to be incorporated into a probabilistic programming context because it can be formulated as a set of transformation rules.

Joint work with Frank van der Meulen (TU Delft). 

Zoom notes: The passcode for this meeting is 321777. This meeting ID — 621 4469 8204  — will be the recurring meeting for the Statistics and Probability Seminar.

Page responsible:webmaster@math.kth.se
Belongs to: Department of Mathematics
Last changed: Oct 29, 2020