Estimation of Langevin Drift and Diffusion Functions from Data
Basic data for this talk
Type of talk: scientific talk
Name der Vortragenden: Willers, Clemens
Date of talk: 14/09/2017
Talk language: English
Information about the event
Name of the event: Workshop Selbstorganisation und Komplexität 2017
Event period: 10/09/2017 - 16/09/2017
Event location: Zaferna Hütte, Mittelberg, Österreich
Organised by: AG Thiele, ITP der WWU Münster
Abstract
There are many examples, in which one wishes to characterize the dynamics of observed data through Langevin equations. In this talk, several methods are presented, which make it possible to estimate the drift and diffusion functions, which define a Langevin equation: the direct estimation, maximum likelihood parameter estimations and the Bayesian approach. For the parameter estimations, one needs a so called propagator, which makes it possible to calculate transition probabilities. It is known for the one dimensional case. Besides, there are data sets, which can only be characterized by Langevin systems of higher dimensions (because of special properties in smoothness and autocorrelation). A possibility to calculate a short-term propagator in the two dimensional case is presented.Keywords: statistical data analysis; nonlinear dynamics; Langevin; estimation methods; maximum likelihood; parameter estimation; Bayes; propagator; machine learning
Speakers from the University of Münster
| Willers, Clemens | FB11 - Faculty of Physics (FB11) |