Master Class: Lecture 2: Maximum Likelihood Particle Parameter Estimation for State-Space Models

SpeakerArnaud Doucet
AffiliationUniversity of Oxford
DateThursday, 04 Oct 2012
Time13:00 - 14:00
LocationRoberts G06 Sir Ambrose Fleming LT
Event seriesMaster Class: Arnaud Doucet (1-5 Oct 2012)

In most applications, the state-space model of interest depends on unknown static parameters that need to be inferred from the data either online or offline.

The aim of this lecture is to explain why many naive and sophisticated particle schemes proposed in this context are unreliable in the presence of large datasets. I will then present original approaches to perform online or offline parameter inference. These provably stable schemes rely on an exact forward only implementation of the celebrated forward backward recursion originating from dynamic programming.

Slides: ucl_2.pdf

iCalendar csml_id_107.ics